我们如何为 AI 重构了 Airtable 的整个组织架构 | Howie Liu(联合创始人兼 CEO)
How we restructured Airtables entire org for AI | Howie Liu (co-founder and CEO)
Howie Liu: If you were literally founding a new company from scratch with the same mission, how would you execute on that mission using a fully AI native approach? If you can’t, then you should find a buyer and then if you really care about this mission, go and start the next carnation of it.
The Viral “Airtable is Dead” Tweet
Lenny Rachitsky: Or people that work for you, how have you adjusted what you expect of them to help them be successful?
Howie Liu: If you want to cancel all your meetings for like a day or for an entire week and just go play around with every AI product you think could be relevant to Airtable, go do it.
CEO Returning to Write Code
Lenny Rachitsky: Of the different functions on our product team PM, engineering design, who has had the most success being more productive with these tools?
Howie Liu: It really does become more about individual attitude. There’s a strong advantage to any of those three roles who can kind of cross over into the other two. As a PM, you need to start looking more like a hybrid PM prototyper, who has some good design sensibilities?
CEO Using AI as a Success Signal
Lenny Rachitsky: Do you see one of these roles being more in trouble than others? Today, my guest is Howie Liu. Howie is the co-founder and CEO of Airtable. I’m having a bunch of conversations on this podcast with founders who are reinventing their decade plus old business in this AI era, to help you navigate this existential transition that every company and product is going through right now. Howie and Airtable’s journey is an incredible example of this, and there’s so much to learn from what Howie shares in this conversation.
We talk about a very interesting trend that I’ve noticed that Howie is very much an example of, of CEOs almost becoming individual contributors again, getting into the code, building things, leading initiatives themselves. That’s something that we call the IC CEO. We also talk about the very specific skills that he believes product managers and product leaders, also engineers and designers need to build to do well in this new world that we’re in. Also, how he restructured his company into two groups, a fast thinking group, and a slow thinking group, which allowed their AI investments to significantly accelerate.
If you’re struggling to figure out how to be successful in this new AI era, this episode is for you. If you enjoy this podcast, don’t forget to subscribe and follow it in your favorite podcasting app or YouTube. Also, if you become an annual subscriber of my newsletter, you get a year free of 15 incredible products, including Lovable, Replit, Bolt and Adyen, Linear, Superhuman, Descript, Wispr Flow, Gamma, Perplexity, Warp, Granola, Magic Patterns, Raycast, ChatPRD and Mobit. Check it out at lennysnewsletter.com and click product pass. With that, I bring you Howie Liu.
Lucidlink fixes this. It gives your team a shared space in the cloud that works like a local drive. Files are instantly accessible from anywhere. No downloading, no syncing, and always up to date. That means producers, editors, designers, and marketers can open massive files in their native apps, work directly from the cloud, and stay aligned wherever they are. Teams at Adobe, Shopify and top creative agencies use LucidLink to keep their content engine running fast and smooth. Try it for free at lucidlink.com/lenny. That’s L-U-C-I-D-L-I-N-K dot com slash Lenny.
To learn more, visit DX’s website at getdx.com/lenny. That’s getdx.com/lenny. Howie, thank you so much for being here and welcome to the podcast.
Biggest User of AI Inference Costs
Howie Liu: I’m so excited. Thank you, Lenny. I’ve been a listener from afar for a while now.
Reshaping Time Allocation
Lenny Rachitsky: I’m really flattered to hear that. I’m also very excited. You’ve been on quite a journey over the last, is it 13 years, is it longer?
Howie Liu: Yeah, right about 13.
Restructuring the Organization
Lenny Rachitsky: 13 years. I imagine there’ve been a lot of ups and a lot of downs. I want to talk about all those things. I want to talk about a lot of the lessons that you’ve learned along the way. I want to start with what I imagine was a very surprising down moment in the history of Airtable. This is something that, unfortunately, something I think about when I think of Airtable. I feel other people may feel this way, is there’s this tweet that went super viral, maybe a couple of years ago at this point where someone just shared all this data and they’re like, Airtable is dead.
They’ve raised way more money than they’re worth. They’re not making enough to get from underwater. Yeah, Airtable RIP. What happened there? How much of that was true? How did that go?
PLG vs Sales-Driven AI Distribution
Howie Liu: Yeah, so basically none of it was true. I mean, the surprising thing to me was how viral this tweet went when … Frankly, I actually look back at this person’s other tweets. I think they worked at CB Insights, and the irony is the whole point of that business is to have good data, good data quality around private company data. And they just literally had incorrect numbers by a strong multiple on what our revenue scale was, what our growth rate was. And if it gave me some consolation, I look back and this person had also tweeted about other companies, like Flexport was the last take-down tweet.
They have like, “Oh, Flexport’s dead” and their evaluation is too high, and blah, blah, blah. And so, I think that the more surprising thing was just like this person has been tweeting a bunch of spicy takes that are not substantiated by real data or correct data, and yet this particular tweet went super viral and that was the perplexing part to me. And then, actually, I think what really gave it legs was on the All In podcast, which is obviously super popular. And I listened to it. They covered it. They were like, “Oh, latest on this week’s news, this tweet about Airtable. What do we think about this?”
And it almost, I think became a way to talk about a broader theme of what happens to this last generation of highly valued companies, maybe decacorn companies in this new … And at that point, it was the recent moment for both public and private markets. They did also issue a correction though. All In, did a follow-up episode, a few, I think weeks later saying, “Hey, we got the numbers wrong. We are revising our case and a view on Airtable.”
Lenny Rachitsky: What’s that line about how a lie gets around the world some number of times before truth has even this time to get out of bed?
Making AI the Default Experience
Howie Liu: Yeah. Well, I think I learned about memes and morality very quickly in that experience. Not a very good social media person, but I think I learned a little more.
Lenny Rachitsky: Yeah, it’s tough. Twitter is such … The incentives are so misaligned. It’s just I tweet something people want to share, not truth.
Vibe Coding and AI-Native Creation
Howie Liu: Well, especially … I mean, there’s a lot to like … I would say, I like the post Elon Twitter more than the pre Elon Twitter because it is just bolder, and I guess I really admire bold product execution where you’re not just stuck to the current laurels and they’ve made so many changes, but I do feel like I get injected into my feed very sensational content all the time, and I mean, it works on me. I can’t help but to click on it and engage with it, but it does … I think it does result in this kind of content, really spreading.
Lenny Rachitsky: Yeah. Now, Nikita running the show, I don’t know if you saw this, there’s a new … We don’t need to keep talking about Twitter, but there’s a new feature where you take a screenshot of a tweet and it has a huge X.com logo watermark on the top, right? Yeah, just to … People are sharing these tweets all the time. Yeah.
Vibe Coding Limits vs Airtable Strengths
Howie Liu: Yeah.
Lenny Rachitsky: Man. Never a dull moment over there.
Completing the [Glitched Text]
Howie Liu: For sure.
Commonalities in the Founder Journey
Lenny Rachitsky: Okay. I want to go in a completely different direction, something that I’m really excited to talk to you about, which is this very emerging trend that I’ve noticed that I feel like you’re at the forefront of CEOs becoming ICs again. It’s kind of this move of, IC CEOs. CEOs getting their hands dirty again, building again, getting in the weeds, coating again. I feel like you’re again at the forefront of this. Talk about just why you’ve done this, why you think this is important, and just what that looks like day to day to you versus what your life was like a few years ago.
Extensively Testing AI Products
Howie Liu: The underlying reason for this shift, at least for me, is that, as we started the company, I was very much in this mode. I was literally writing code both on the backend, thinking about the real time data architecture of our platform, also the front end, the UX. And I would argue that in that founding moment, the initial product market fit finding, and especially for a product that is pure software, we weren’t building an operationally heavy business like a dog walking marketplace where the tech is only an afterthought.
The tech was the product, right? And in a very Meta sense, Airtable is the platform for other people to build their own apps. So it’s all about the attack, like the very intimate design decisions, again, both architecturally and on the front end and the product UX choices. That is the product’s value prop. You can’t separate those two. You can’t say, “Okay, I researched the jobs to be done. Here’s the workflow, here’s the process, and then, okay, some engineer can just build it as an afterthought.”
It’s those little decisions and really be able to be at the bleeding edge of what’s possible both in the browser and with the real-time data architecture. That made the product what it was. I think the same is true for Figma, which actually had a very parallel timeline to us. We both were founded around the same time, both spent two and a half years building the product, hands-on that early team before launching. And when I think now to both the era in between that founding moment and then now as well as now the new gen AI moment, I think there was a maturing era of both SaaS overall and Airtable specifically.
Where, as you scale up and you learn how to build teams and organizations and you have to scale up stuff that’s not actually those intimate details, but process and people and so on, you kind of get by default further and further away from those details, right? And maybe for some businesses that’s fine because no longer is it about finding the details that make for a magical new product market fit. And it is really just about scaling up an existing thing that works and using what I would call more blunt instruments to scale it up, like a more blunt roadmap, a more blunt go-to market execution strategy.
Regardless, I think that now, we’re entering this moment where … Certainly every software product in my opinion, has to be refounded because AI is such a paradigm shift, it’s not even just like the shift from desktop to mobile or on-prem to cloud where that was more like a very one time and somewhat predictable change in form factor. I think AI is so rapidly evolving that with every evolution, every new model release and every new type of capability that’s released, it actually implies novel form factors and novel UX patterns to be invented to fully capitalize on those capabilities.
And so to be continuously relevant and to refine product market fit in this era, I think you have to be of the details. There is no looking at it from 10,000 foot view and saying, “Oh, we’re just going to throw a bunch of people at this problem.” It’s actually understanding what is the right product experience and the right business model that backs it up and the right … everything else to support that engine to take advantage of the capabilities in our product domain.
Encouraging the Team to Play with AI
Lenny Rachitsky: You have this phrase somewhere where you talk about being the chief taste maker.
Prototypes Over Slide Decks
Howie Liu: Yeah.
Who Benefits Most in Product Teams
Lenny Rachitsky: And to do that, you have to do exactly what you’re describing.
Product Taste Comes From Hands-On Practice
Howie Liu: That’s right. I mean, I think that, and I would also say it’s actually now also hard to taste the soup without participating in at least some part of creating the soup. Meaning With AI, you can kind of look at the final product and say, “Okay, this feels right or not, or it feels like we’re being bold enough and we’re properly productizing these new capabilities.” But I think to really understand the solution space of what’s possible, you have to be in the details.
I mean, literally, you can’t just look at screenshots or a pre-recorded video of a new product feature. AI is something you have to play with, and ideally you’re playing with both the packaged up app or solution that you’ve built with it, but you’re also playing around directly with the underlying primitives who are using the models either via API or via a chat interface. You’re really pushing them to the boundaries. Because that’s the only way that you really understand what these new ingredients. It’s like as a chef, you just gained access to amazing new ingredients, but you have to actually get comfortable with them to put them into a new dish.
Lenny Rachitsky: And we had Dan Shipper on the podcast, he runs this newsletter and podcast to product a company called Every. And they work with companies to help them become more AI successful and adopt AI and all that stuff. And I asked him, what’s the signal that a company will have success adopting AI and seeing huge productivity gains? And he said, does the CEO use ChatGPT or Claude daily?
Evals and Early Product Exploration
Howie Liu: Yeah.
Lenny Rachitsky: And I feel like you’re describing exactly, hourly,
Evals Are Better for Iterative Improvement
Howie Liu: Literally hourly, or you could even have a measure of inference costs, right? Like the equivalent underlying inference compute cycles, right?
Key Advice for AI Transformation
Lenny Rachitsky: How many tokens you use?
Breaking Silos: Everyone Goes Full-Stack
Howie Liu: Yeah, I mean, I’m proud to say I am pretty sure I’m still the … I just checked this recently, but I take pride in being the number one most expensive in inference cost user of Airtable AI, not just within our own company, but I think for a long time I was globally across all our customers vault. I mean, I’m extremely intentionally wasteful. Wasteful in the sense of I’ll do something that costs maybe hundreds of dollars of actual inference costs. For instance, doing a lot of LLM calls against long transcripts of let’s say, sales calls to extract different types of insights like here’s the product apps, identify or here’s summaries, et cetera.
And we also have now a capability that’s basically like an LLM map reduce. So effectively, even if you can’t fit the entire corpus of content into one LLM call, because the context window limitations, we’ll map through all of this content and break it up into chunks and then perform an LLM call on each one and then perform an aggregation LLM call on those chunks. Very expensive, because you’re basically running a highly expensive model against a lot of data and then running it again on the aggregates of that. But for me, hundreds of dollars spent on this exercise is trivial compared to the potential strategic value of having better insights.
It’s as if a really, really smart chief of staff has gone through and read every single sales call transcript that we’ve had in the past year and giving me very astute product insights, marketing insights, kind of positioning insights and segmentation insights. That’s invaluable. You could pay a consulting firm literally millions of dollars to get that quality of work. So to me, I still think the value versus the actual cost of AI when applied greedily but smartly, it’s a crazy ratio. And more people should be aggressively throwing compute cycles at these very high value problems.
Most Counterintuitive Startup Lessons
Lenny Rachitsky: Until somebody tweets how you’re costing the company so much on AI compute and you guys are going to be underwater.
Product Lovers vs Opportunity Hunters
Howie Liu: I’m just kidding. It’s like how we have personally taken down the cashflow profile of the business.
Lenny Rachitsky: So CEO’s, founders hearing this, they’re probably like, okay, I should probably start doing this. What does this actually look like? I imagine you still have a lot of other stuff you got going on once, you got all these … How do you change your day to day to do this?
Everyone Can Be an All-Around Builder
Howie Liu: Yeah, so I actually cut my one-on-one roster by default, and the idea is not that I don’t want to spend time one-on-one with people, but rather that I found that the … Just having more standing one-on-ones actually precludes me from engaging in more timely topics. I like to think of the best types of meetings as very urgency driven. And there’s some timely topic, you’ve discovered some insight. Maybe I talked to some new startup and I learned something from their product or their approach.
And I want to bring that into how we’re thinking about a new feature at Airtable or even just plant the seed with some different EPD people within Airtable, I want to make most meetings very timely and very informed by real alpha. There’s got to be some kind of value and insight to seed that with. Now, in addition to that, I’ll supplement with, when I’m in person with someone, I want to carve out time for a proper catch up and less structured, less timely, and just more of building a relationship with a human.
But I actually find that having that common .. It’s almost a barbell approach where it’s like if you’re going to spend time with somebody in a freeform way, actually do it in a high quality, not forced weekly ritual way. Go for a longer lunch or coffee walk or whatever in person when you can. Maybe that’s a once every month or two kind of thing. And then the in-betweens are either topical, so we do have standing meetings for … Now, we have a weekly basically sprint check-in on all of our AI execution stuff, which now is half the company or half the EPD org is working on AI capabilities.
We’re trying to ship very quickly, like I basically want to always ask the question, how would an AI native company, like a cursor or windsurf, et cetera, how would they execute? And are we executing as fast as them and taking advantage of all the new stuff as well as them? So bringing that level of intensity and urgency to how I spend my time within, that’s been the biggest shift for me.
Lenny Rachitsky: What’s a change you’ve made to help the company move faster and match that sort of pace?
The Best Era to Learn Tech
Howie Liu: Yeah. I mean, we did do a reorg of the EPD org. So before we had … we’ve gone through a few different reorgs over the past, call it, four years. The original state as we just proliferated, I think by default or incrementally, was that we had a bunch of groups that were each responsible for a feature or a surface area. So there was a group responsible for search within our table, and there was a group responsible for mobile experience and so on and so forth. And that has its benefits. Obviously, that team can go and get really ramped up on that part of the code base, that part of the product.
But it has the disadvantage of yeah, you tend to think incrementally when everyone’s remit is actually a feature that they incrementally improve by definition as opposed to thinking about a mission or a outcome goal that might need to coordinate dramatic changes across a wider set of surface areas instead of just each one incrementally improving. And so, we reorged initially to basically different business units effectively. So I know Airbnb has done the functional to GM back, et cetera. This was more like saying, “Look, we have an enterprise business” and them MO there is more about scalability.
Can we support the larger scale data sets and use cases? Do you have the core capabilities needed to be able to push out an app to maybe 10,000 seats or 20,000 seats for product operations? A lot of architecture, a lot of scale, that kind of work. We would have, what we call the teams filler, which is more about self-serve, kind of the product UX, how easy it is to adopt the product on board, share, do all the kind of basic functionality. An AI pillar, solutions pillar, and basically infra. And what we found though with that approach is that there was still … there was more kind of holistic bets being made.
So the team’s pillar could think not just about one feature, but the overall onboarding experience where really about Nuxt in a way that touched multiple parts of the product, but it still felt like it wasn’t … Especially as we started to execute more on AI stuff, it wasn’t allowing us to aggressively and quickly move as a AI native company would. I mean, when you look at the cursors of the world, they’re shipping major new stuff every week. And it’s not like, “Oh, well we have this separate roadmap for enterprise, we have this roadmap for this group.”
And it just feels like one cohesive product that’s shipping at a breakneck pace. So we did this recent reorg where now we have what I call the fast thinking group, which officially is called AI platform, but it really means we want to just ship a bunch of new capabilities on a near weekly basis. And each of them should be truly awesome value. You should drop your jaw, how awesome it is to use this new capability in Airtable. And then separately, we have the slow thinking group, and that’s not meant to be better or worse. It’s literally like you need fast and slow thinking in the common sense to operate as a human.
Rapid Fire Q&A
Lenny Rachitsky: I have that book behind me.
Howie Liu: Yeah, I love that book. But slow thinking it’s like, it’s just a different mode of planning and executing, right? It’s like more deliberate that require more premeditation. We can’t just ship a new piece of infrastructure that has a lot of data complexity like our data store HyperDB that now can handle multi-hundred million record data sets. That’s not something you ship in a week in a hacky prototype. So we now have these two separate parts of the company, and I actually think what’s really cool is they actually compliment each other very well, right?
Because the fast execution, the AI stuff, that creates the top of funnel excitement that also inspires new use cases and new users to come to Airtable, including in large enterprise, right? Enterprises can use this stuff too. It’s not just like a SMB thing, but the slow thinking basically allows those initial seeds of adoption to Sprout and grow into much larger deployments. Whereas I think a lot of the challenge for many of the AI native companies I’ve seen is that they could have a very wide top of funnel, like get all of this AI, tourist traffic.
A lot of interest, a lot of early usage, but then sometimes the challenge is how do you turn that into more durable growth and get each of those adoption seeds to retain and expand over time.
Great Products Discovered Recently
Lenny Rachitsky: That is super cool. I’ve never heard of this way of structuring teams, the fast thinking, thinking fast, thinking slow, the Kahneman. It’s so interesting for the fast thinking team, do you find there’s specific archetypes of people that are successful there? Is it a lot of bringing in new people that are not just used to the way of working at our table? What do you find?
Howie Liu: We have a mix. So we brought in … I mean, we’re always hiring, right? There was never a point in the company’s life where we stopped hiring. And candidly, even when we had to do two rifts, that’s significantly reduced our head count. We had just way too quickly grown and overscaled the business at a certain point. But even when we did our rifts, we were still actively recruiting and hiring in … I mean every major department, but especially in EPD, because it’s always been my belief that it would be arrogant to say that we have all the people we ever need already in the roster today, right?
We’re always going to need to find new, fresh perspectives, new skillsets, et cetera. And so, we’ve continued to hire … I think we’ve learned as we’ve gone along of what is the ideal type of hire, and we’ve done some actual hires and learned from that as well. But I think the fast thinking part, it really just requires a lot of … Somebody who’s able to operate with a lot of autonomy, who’s entrepreneurial in nature. Now, it doesn’t mean they have to literally be a former founder. I know some companies are, like Rippling for instance, does a lot of actual acquisitions and gets actual founders into the company.
We found that that’s great and we’ve done some of that as well. But also there are some really, really capable people who we didn’t literally have to acquire in, and yet, they’re just able to think full stack about the problem and the user experience. Problem, not just meaning the technical layers of the problem, but also, what is the wow factor we’re trying to create. So tangibly we’re doing this new thing that’s about to ship, where not only can you describe the app you want to build and then iterate on it with our conversational agent Omni.
And it builds it with the existing air table platform capabilities, but we’re also giving it the ability to actually do code gen, to extend those apps with really final mile very bespoke functionality or visuals. So you could say, “Hey, generate me a very, very specific type of map view with this kind of heat mapping and this kind of icons and …”
It’s kind of like heat mapping and this kind of icons. And when you click it, do this. And that’s a capability that there’s so much ambiguity in some of the design decisions around it. And you have to blend that design thinking with some of the technical constraints of what can the AI models actually one shot effectively?
And if not, how do you add in the right human workflow for approval and review, and the reprompting and so on? So just so many different design decisions, and you need somebody who can really think full-stack about that kind of product and is not overwhelmed by that kind of open-endedness, but relishes in it.
Words to Live By
Lenny Rachitsky: I was actually playing with it before we started chatting. I made a really cute startup CRM.
Howie Liu: Oh, that’s awesome.
Lenny Rachitsky: Yeah, started talking Omni over here. It’s like the colors are beautiful-
Howie Liu: [inaudible 00:26:47].
Lenny Rachitsky: … so that’s what’s standing out to me right now.
Howie Liu: [inaudible 00:26:49] there is…
Lenny Rachitsky: Yeah.
Howie Liu: I will say just as a note, I consider myself at my core a product UX person. That’s my passion. And everything else I’ve had to learn to run this company is almost like what was a necessary part of the journey. But my real passion is thinking about product UX. And I think of UX in a deeper sense than just the cosmetic design. What you could put into a framer kind of prototype. I think of it as literally what should this product do and how should it represent that and behave for the user? That is the product, in my opinion, right. And of course, then you have to figure out technically what’s possible and how to implement it.
But I think to me what’s under executed today in the world of AI products is there’s so many awesome capabilities of AI, and most of them are really under merchandise, and there’s very poor, actually, visual or otherwise metaphors or affordances given to users to help represent or understand what those underlying capabilities are. I mean, ChatGPT obviously extremely successful product, so not knocking it at all, but you come in and you just get this completely blank chat box by default, and now they have suggestions underneath and so on.
But the product UX part of me is just craving more visual metaphors or colors or some kind of use the canvas of a web interface and all the richness interaction you create there to better represent or show all the different things that you can do with the underlying model, right. And so that’s something we’ve tried to do with Airtable, is show all of the different states and use colors even to play those up.
Lenny Rachitsky: It’s interesting how much of this connects with I just had Nick Turley on the podcast. He’s head of ChatGPT at OpenAI, and he had these two really interesting insights that resonate directly with what you’re describing. One is he has this concept of whenever something is being worked on, he’s always asking, ” Is this maximally accelerated? How do we move faster? If this is important, what would allow us to move faster?”
Howie Liu: Yeah.
Lenny Rachitsky: And I love that that’s one of the themes that’s coming up as you talk, is just this creating this very clear sense of speed. And you even call it the fast-thinking team, like, “You are going to move fast.” And then the other one is just this insight that with AI, you often don’t know what it can do and what people want to do with it until it’s out. So there’s this need to get it out, and that’ll tell you what it should be.
Howie Liu: I couldn’t agree more with both of those, and particularly on the second point, I think it’s interesting. Clearly, there have been companies that have both been successful in PLG and more sales-led distribution for AI products. The most notable ones I can think of are Palantir with their AIP deployments. That’s obviously very sales-led. You’re not PLG into a Palantir deployment. But even companies like Harvey and so on, they’re doing very well. And it’s primarily, from what I understand, sales-led.
You’re not self- serving into a Harvey instance at a law firm. And yet, to me, the best way to get AI value out there is experientially, right. And so you can kind of get that in a sales motion. You can show a demo. Maybe you can do a POC, but it’s so much more powerful when you just open up the doors and say, “Anyone who wants to come and sign up and trial this product can.” And I think to me, it’s a real proof point that ChatGPT is arguably the most successful kind of PLG product of all time, just in terms of sheer scale of users. Like they announced 700 million… Is it MAUs or week… I think it’s actually-
Lenny Rachitsky: Weekly active users.
Howie Liu: Weekly.
Lenny Rachitsky: 10% of humans on earth use it-
Howie Liu: That’s insane.
Lenny Rachitsky: … weekly.
Howie Liu: That’s insane. In how many years? A few years.
Lenny Rachitsky: Three years. Under three years.
Howie Liu: Yeah. So I mean, literally, that is just the most insane ramp curve. And I don’t think they would’ve gotten there if you couldn’t just come in and literally try the product out. And as a little bit of a rebuttal of the point I made earlier where I think ChatGPT doesn’t do a ton right now, and even earlier they did even less to expose all the different ways you could use it, but they just made it so frictionless to just try it for yourself that you as a user could come in and just literally ask it anything and see how it did. And of course, people in the early days tried to stump it and showed, “Oh look, see, it’s not that smart. It doesn’t answer this hard question really well.”
But clearly, the magical nature of it still appealed to you enough. Everybody used it. And so I think I do have a view. We’ve gone through that whole arc of we started PLG. I’d like to think Airtable was one of the PLG darlings of our era. And anyway, I started moving up market and doing more sales execution, although that was still always on top of usually PLG within an enterprise, but we started doing more and more sales execution. We still have that. That’s still really important for our business. But I also think, me personally, one of my goals is to shift my attention back into that kind of builder-led adoption and literally showing in the product experientially, not telling in a deck, the value that you can get from AI and Airtable.
I think that’s so key, and it’s [inaudible 00:32:23], but it’s also more than that. It’s not just literally how do you onboard somebody into the product. It’s literally thinking about the entire product experience itself, right. And in our case, we just made the entire product experience AI-centric. It used to be that we had kind of this secondary thing that you could ask questions to the assistant sidebar. We now made our agent the default way of doing everything in Airtable, and it’s now the Airtable app, as you know, it is almost like an artifact that’s manipulated by and can be tool used by the agent.
Lenny Rachitsky: Let me follow that thread. So if you go to Airtable.com today, it looks like basically all the other AI app building sites. Now it’s just tell me what you want to build. Thoughts on that, as just a thing everyone’s starting to do is there… what do you think comes next? Is this… Is it working well?
Howie Liu: There’s clearly an incredible magic to vibe coding and app building with AI. And this is actually a prime illustration in my view of that concept we talked about a second ago, which is as capabilities of these underlying models evolve, the form factor in the product UX also needs to evolve with it. And so the earliest models, like the kind of original ChatGPT, like GPT-3.5 kind of era models were not nearly as smart as the current models. And so you couldn’t really ask it to one shot a more complicated chunk of code, or certainly not like a full stack app, and expect it to work.
And so the right form factor for leveraging those models in a software creation context was GitHub Copilot, right. It’s like auto-complete a few lines of code at a time. But you couldn’t chat to it and tell it, “Build me this entire app from scratch.” And I think that as the models got better and better, you saw that the new form factors emerge. I think Cursor did a great job of being an early pioneer of this more age agentic way of leveraging the models to do more complex things and generate more larger chunks of code.
And now with Composer, you can literally just go into Cursor and build an app from scratch, build me a 3D shooter game from scratch, and just watch it go and create all the files and fill out each file, and then the thing actually runs some of the time. And so to me, this is where the world is going. The models are clearly getting smarter. And if you think about the original vision of Airtable, it was always about democratizing software creation. We just strongly believed that the number of people who use apps far outweighs the number of people who can actually build their own or manipulate apps and harness custom software to their advantage.
Lenny Rachitsky: That sounds very familiar, very familiar these days.
Howie Liu: Yeah, exactly. And so I think this is, it’s a different means to the same end. And so it’s almost like we have to lean into this because if we started Airtable today, this is what we would be all in on. Now I think that the advantage that we have, and I do think you have to be realistic to yourself, especially as a company that predates GenAI and now has to find your new footing in the AI landscape. You can’t fool yourself and just say like, “Okay, I’m going to throw in some AI stuff on the landing… on the marketing site, put in a couple AI features, and call it a day.”
I think you actually have to take a clean slate approach to saying, “How would our mission best be expressed? If you were literally founding a new company from scratch with the same mission, how would you execute on that mission using a fully AI native approach?” And then, by the way, do you have useful building blocks that you can leverage from your existing product and your existing business, or are you literally worse off having this legacy asset versus starting something from scratch? And I don’t think the answer is always yes or no. I think it just depends on the product.
And if you can’t really introspect and say, “Look, I think I’m better off doing this with the pieces that I have for my existing business and product,” then I think you should sell. You should find a buyer for that company and then go. And if you really care about this mission, go and start the next carnation of it. In my case, I really thought about this and really feel strongly that the building blocks that we have, these no code components, actually do allow us to execute better on this vision than if I had to start from scratch.
Meaning the problem with vibe coding, especially if we’re building business apps… So I should clarify that we want to democratize software creation, but specifically, we are focused on business apps. We’re not trying to be the platform where you create a cool viral consumer game. This is for like your CRM, right. Or if you want to build an inventory management system as a small restaurant or a lawyer trying to build a case management system, that’s what we’ve always been focused on. And I think in this AI-native world, clearly, you should be able to generate those apps agentically.
And yet if you have an agent that has to generate every single bit of that app from scratch, from code, it’s going to be very unreliable. There’s going to be bugs. There’s going to be data and security issues. And then you’re also going to have a context collapse, as it just cannot manage all of the code that it’s written, basically, as the app gets more and more complex. And what we actually have are basically these primitives that the agent can manipulate and use without having to literally write the code from scratch to represent, “Here’s a beautiful crud interface on top of the data layer.
Ours is real-time and collaborative, and really rich, and has collaboration on it. And by the way, here’s all these other view types and a layout engine for a custom interface, a layout, or automations and business logic.” And so it’s almost like in programming terms, the Airtable pieces in our Lego kit today can be used by this agent as almost like a more expressive DSL, like a domain-specific language to build business apps instead of literally having to write everything down to the SQL and HTML and JavaScript to build every part of that app from scratch.
And so if we can combine the best of both worlds, we have these very reliable, high-quality Lego pieces. Now, an agent can go and assemble them for you instead of you just using the GUI to do that. And by the way, if you do want to fall back to the GUI, there’s a really great kind of way for the non-technical user to still understand and participate in what’s going on. Whereas if you’re not technical, you can’t inspect the code underneath a v0 or Lovable or Revolut app, right.
It’s just kind of opaque to you. And if you can’t re- prop it to get what you want, you’re kind of stuck. This is much more akin to a developer using Cursor can generate lots of code, but then can still drop back to the IDE to edit and manipulate it to the final production-ready state. So that’s kind of the play that we’re making. And if I didn’t fully and truly believe we have a better shot at doing it with our existing product, I wouldn’t be running this company in its form today.
Lenny Rachitsky: I’m talking to a lot of founders that are going through the journey are going on, which is, “We’ve had a business for a decade, AI emerged, and wow, we got to figure out something that works… that could work even better.” And so I’m trying to pull out the threads that are consistently working across these journeys because I think a lot of companies are trying to figure this out. So one that you just touched on is just if you were to start today, what will you do?
What would that business be? Plus, how can… do we have an unfair advantage with the thing we’ve done in the past? That feels like an important ingredient. And then the other… circling back to stuff you’ve shared already, there’s just creating a sense of urgency and pace and getting people to understand this is how things move in AI, and we need to create this fast-thinking team. I love that metaphor in framing.
And then there’s the point you made about just talking to AI regularly as the founder feels like an important element, just like to truly be this ICCO talking to AI, working with AI regularly. Just on that note a little bit more, just to give people a sense of what this looks like day to day. So you’re talking to Omni all day trying to and undertook… flex the power of what you can do and iterate on it. Is there anything else you’re doing day to day that helps you figure out what to do for the business?
Howie Liu: One, I try to use as many different AI products, including not Airtable, as I can, and both literally for the novelty factor and just some new cool demo comes out. Like Runway release their immersive world engine, and so I’m going to go try it out. When Sesame AI put out their cool interactive voice chat demo, I tried that out because even though we don’t have a direct and near-term need for really realistic and interruptible voice mode where it’s not as core to our capabilities, I just want to understand and get a feel for everything that’s out there.
And I try to invent little, almost like side projects of my own, to have a real reason to use these products. Like, “Oh, cool. What if I were to take… What if I were to try to create a funny little short… a funny video short using a combination of HeyGen avatars with a script, like a comical script generated by AI? And maybe it’ll be on an interesting topic. So I’ll do deep research on the topic with ChatGPT and pull together the results, have it compose kind of a little dialogue.
Lenny Rachitsky: Did you actually do this? Is there something you made?
Howie Liu: Yeah. That’s literally an example of something, just a fun weekend project. And to be honest, these things only take you an hour if you become kind of pretty proficient with using the products. They’re all so easy to use. You can literally do the deep research thing, kick off query, make a coffee, come back in 20 minutes. Okay, let me prompt it to generate me some dialogue. It’s a little bit like what NotebookLM does for you out of the box, but sometimes I like to just do it myself. And then, okay, let me take the script and cut it up and turn it into a HeyGen avatar and then download the video and play it.
And just for fun. I’m not trying to make that into an actual YouTube video business. But I think coming up with these different fun weekend projects is a really useful construct to force myself to actually try these products in a more than just a Twitch click way. And what it gives me is, A, it’s not just understanding the models, which is also very, very important, right. GPT-5 came out yesterday, and playing around with it a bunch just on a variety of different personal use cases, but there’s a difference between just understanding the model but then also understanding the product form factors in which they can be placed, right.
Meaning when you apply the model in a more structured way, when you apply the model with different tool calling than maybe what ChatGPT has in its out-of-the-box form, when you apply it with a more agentic workflow, again, that might be different from what ChatGPT gives you out-of-the-box, that’s when you kind of learn you really get to inspire yourself on what are the product’s form factors that these new models can take. And plus, by the way, I find it to be really fun. There is to me a delight and entertainment value to just using AI, period, because A, it’s not perfectly predictable.
So I think the element of you’re not quite sure what you’re going to get. It’s like a box of chocolates. And B, it always blows my mind just to think about, “Wow, five years ago we didn’t have any of this stuff.” AI was like, okay, it’s like we can do predictive analytics. There’s some basically very advanced kind of regressions that we could run with AI, but it looked nothing like this in its current form, and it’s just actually super fun, in my opinion, to get to play around with all the different types of products that come out.
I think that is a big part of it because on the point about the pace of the world moving so much faster in AI than any other landscape in SaaS, in the mature SaaS era, it was important to study your competition. If you were building a SaaS company, you’d be crazy not to follow Salesforce every year and see what the major releases they’re putting out are, or ServiceNow, or so on.
This is the equivalent of that, but there’s major new releases and products and so on every week, not every year. And so I just think you have to say abreast of all… of it all and combining this with our point earlier of a lot of this has to be experienced, not just read. You can’t just read the write-up on TechCrunch or even a tweet about a new capability. You kind of have to try it to really get a sense of what it is.
Lenny Rachitsky:
Just last week, I was preparing for an interview with a very fancy guest, and I had Claude tell me what are all the questions that other podcast hosts have asked this guest so that I don’t ask them these questions. How much time do you spend every week trying to synthesize all of your user research insights, support tickets, sales calls, experiment results, and competitive intel? Claude can handle incredibly complex multistep work.
You can throw a hundred-page strategy document at it and ask it for insights, or you can dump all your user research and ask it to find patterns. With Claude 4 and the new integrations, including Claude 4 Opus, the world’s best coding model, you get voice conversations, advanced research capabilities, direct Google Workspace integration, and now MCP connections to your custom tools and data sources.
Claude just becomes part of your workflow. If you want to try it out, get started at Claude.ai/Lenny. And using this link, you get an incredible 50% off your first three months of the pro plan. That’s Claude.ai/Lenny. For people that work for you across Airtable, say the product team, PMs, maybe engineers, designers, how have you adjusted what you expect of them to help them be successful in this new world?
Howie Liu: One is really, really, really stressing this idea of go play with this stuff. And I mean, when I say play, I really mean play in the psychological sense of there’s a difference when you go in and you’re kind of just trying to check the box and get a job done. There’s a difference when you come in with a curiosity and you’re kind of exploring, right. And it’s both more fun and energizing, but also, I think you learn more through that. And so I’ve really tried to stress the value of play with these AI products.
And I kind of try to lead by example, by literally going and sharing out links or screenshots of the things that I’m doing in these various products. So, as an example, I will go into one of the prototyping tools and show, “Hey, I built a marketing landing page for this new capability we’re launching.” I created a landing page for it in Replit, let’s say, and now I’m sharing that link. Instead of what typically we would’ve done in the past is like, okay, we’re going to write a doc about it and then share the doc, I’m just going to show you an actual landing page with visuals and everything in there.
Or I’ll share the actual link to my deep research reports. Or instead of me writing a perfect memo on a topic, I’ll actually just prompt my way into getting a chat thread or a chat output that basically covers all the content that I care about and maybe even ask it to, “Okay, summarize this all into a final memo output,” and then intentionally share that rather than expose the fact that I’m using AI in this way and here’s literally how I’m prompting it so you can follow along as well.
But really trying to encourage everyone to go and just play with these products. And I’ve even said, “Look, if anyone wants to just literally block out a day or frankly even a week and have the ultimate excuse, you could use… you could say that I told you to do it, right. If you want to cancel all your meetings for a day or for an entire week and just go play around with every product, AI product that you can find that you think could be relevant to Airtable, go do it. Period. So I think that’s the most important thing is this play, this experimentation.
I think there’s also a lot of other kind of shifts in how we execute prototypes over decks. I want to see actual interactive demos because, again, it’s hard to… In a deck or in a PRD, you could say, “Okay. Well, we’re going to make Omni really good at handling this kind of app building.” Okay, those are just words. The real proof is in the pudding of like, “Okay, let me try it out on a few realistic prompts that I can imagine.”
And in a demo, in a real prototype, you can instantly try it out on unrealistic rather than golden pathy scenarios and see how it feels too. Does it feel too slow? Do we need to expose more of the reasoning or steps that are happening behind the scenes? Create a progress bar or something like that. But it’s really hard to get that feel of the product with anything but a functional prototype that really does, in an open-end way, use the AI to do whatever you put in.
So I think it’s more like a experimentation playground it feels like how we need to execute, versus I think, in the past, it sometimes felt like a more deterministic resourcing and kind of timelines view of execution. We’re going to put this many people on this problem, and this is the eight-week timeline to this milestone, and then we’re going to ship in a quarter from now. And I think now the whole thing is just a lot more experimentation and iteration-driven.
Lenny Rachitsky: Of the different functions on a product team, PM, engineering design, who has had the most success being more productive with these tools, and how do you think this will impact each of these three functions over time?
Howie Liu: What I found is that it really does become more about individual attitude and maybe some polymathism. There’s a strong advantage to any of those three roles who can kind of cross over into the other two, like the hybrid unicorn types.
So if you’re a designer who can be just technical enough to kind of be dangerous and understand a little bit of how these models work and how does tool calling work and all of this stuff, then you can actually design a concept or even prototype a concept, including in these prototyping tools that’s much more interesting and maybe realistic than if you’re just stuck in the flat like let me put something in a static design concept because I think designs have to be more interactive. The whole… The value of the product and the [inaudible 00:52:04]-
… the value of the product and the product functionality is in the interaction of it, right? Think about the design of Chachi Petite. Again, it’s the most basic design you could possibly imagine. The real design actually is happening underneath the hood in how it responds to different queries and what happens after you fire off a prompt, right? So I think I found that there are people within each of these functions, there are engineers who are very good at thinking about product and experience and can go and prototype out the whole thing. They’re designers who can do the same. Even if they can’t literally code, they can prototype something out literally using a prototyping tool.
I think that’s where AI tooling is also giving more advantage to people who can think in this way by equipping them with an alternative to actually having to go through the long hoops of learning CS, right? Then PMs as well. I think there are some PMs who are really getting into the technical details and studying up on how does this stuff work and actually getting hands-on, rather than seeing the role as writing documents, writing PRDs.
Lenny Rachitsky: Do you see one of these roles, I don’t know, being more in trouble than others, just like you need fewer of these people in the future potentially?
Howie Liu: I think overall you can get more done with fewer people, and that’s not to say we want to go and make the team smaller, but rather … the really cool thing for us and I think a lot of other companies is it’s not like you have a finite set of things you need to do and execute on from a product standpoint, and okay, now I can do that with a 10th of people. I mean, you could do that in a lot of cases, but for us, maybe it’s also because we’re a very meta product, right? We are the app platform with which you can build now any AI app with AI, right? The apps themselves leverage AI capabilities at runtime, whether it’s to generate imagery for a creative production workflow or leveraging deep research, or AI-based crawling of the web to search for companies that match a certain criteria for your Dealflow app or something like that.
We can effectively leverage all of these other AI capabilities in this app platform, because by definition we’re enabling our customers to build apps that have this wide range of AI capabilities. But because of that, it’s like we have a almost infinite set of possible AI capabilities that we could execute on, right? I’m always telling the team like, “Look, the great news is it’s like we have all these fruit trees and there’s so many crazy low-hanging fruit, and you’ve got literally massive watermelons literally sitting on the ground and all you have to do is walk over 20 feet and pick it up instead of having to climb the really tall coconut tree to grab a hard coconut from 50 feet up. So there’s so many watermelons on the ground, just go out and start finding the biggest ones and attacking those, right?”
What that means is that if we can build this culture, and I do think it’s a learnable way of operating, I really like to believe in the growth potential of any human and any individual. I think if you really have a growth mindset, and that’s why one of our most important core values is growth mindset, right? If you really have that growth mindset, I think especially if you’re willing to put in the nights and weekends hours, or in my case I’m literally telling people like, take a full day off, take a full week off and learn this stuff, you can become more fluent in this way. I think then what we get is a team that can just go and work on more things in a much more leveraged and fast way, right?
So, I like to think people who are willing to jump on the train are just going to become more and more effective. It’s not like, oh, as a PM my role is becoming entirely irrelevant, right? No, it means that as a PM you need to start looking more like a hybrid PM prototyper who has some good design sensibilities. By the way, I think some of the best eng PM and design cultures respectively over the past even few decades have always been multidisciplinary in nature, right? The original PM spec at Google required the PMs to actually be somewhat technical so they could understand the engineering limitations of the product designs they wanted to make, and they had to be kind of designy, right?
I remember my co-founder, Andrew, when he was in the APM program was always reading books about design, even down to visual design and color theory and that kind of thing, right? So I think it’s just a reminder that designers as well, some of the best designers through designer to Apple, including hardware designer, you have to understand some of the technical capabilities of how this stuff works, right? If you’re an engineer, I think some of the best engineers and maybe Stripe always had a very good engineering culture of engineers who could think about the product and business requirements. In fact, on any given product group, at Stripe my understanding is that the DRI isn’t always the PM as is traditionally the case in that triangle. Sometimes it’s actually the engineer who’s taking the product lead and saying, this is what we need to build.
Lenny Rachitsky: So, what I’m hearing is essentially the trend across product engineering design is each of those functions needs to get good at one of the other functions at least.
Howie Liu: Yeah.
Lenny Rachitsky: Ideally you can do them all, but if you can just do one additional, so a PM becomes better at design, an engineer becomes better at product management.
Howie Liu: Well, I would actually go further and say I think you need to get decently good at all three. There’s just a minimum baseline of if you’re any one of those roles, you need to be minimally good at the other two, and then you can go deeper into your own specialty, right? You could be a designer who’s really good at thinking about UX and interaction design, and then just good enough to be dangerous on thinking about what’s technically possible and what is the product story around this feature.
Lenny Rachitsky: I love that. To do that, one piece of advice that comes up again and again in what you’ve been describing is use the tools constantly to see what’s possible, and that will teach you a lot of these things.
Howie Liu: I think, well, use the tools gives you exposure to what’s possible, right? It’s kind of like if you wanted to be a great industrial designer, and let’s say, I mean, the chair is the ultimate hello world of industrial design, it’s the canonical design object, you wouldn’t just sit there in a vacuum with no familiarity with the materials that you can use, plywood, steel, whatever, or existing form factors of chairs trying to invent the world’s best chair in a vacuum, right? You should go and first do a study of all of the best chairs out there today. Go look at an Eames chair, sit in it and try to examine it to reverse engineer how it was made, and just look at the prior art for that type of product. That’s how I see the go out and play with these products, and also, I think actually going and designing or implementing or executing is the best practice.
So you can’t just only go and look at other people’s chairs, eventually you have to go and actually try building your own and then try building another one and another one and another one. So, I think that’s where … when I think about how I hone my own product UX sensibilities, I never … and at that time that I was in school and then learning about this stuff, there wasn’t really any good curriculum for UX, right? It’s not like there were great college classes to learn product UX. I mean, even CS was very academic in nature at that time, it wasn’t applied software engineering, like build an app or whatever. Maybe now at some of the schools like Stanford, MIT, they have actually UXy type courses, but it’s still a rarity for most people to have access to that.
So, the way I learned all of my product sensibilities was just trial and error and also using and studying other products, and then going and trying to build my own weekend project ideas, right? Oh, I want to build a Yelp style app with a map view and then also a list view, and I want it so that when you pan around in the map for it to automatically update the list view. Maybe there’s some UX improvements I can make on top of that, but I can also test my technical skills to figure out which parts of this are hard to implement and how do you make it work, and what are some of the design changes or affordances that you can use to map to the technical possibilities.
Lenny Rachitsky: To do that, I loved your piece of advice, which I forgot to double down on, which I also find really powerful. The best tip there is find something to actually build that is useful to you and fun. Pick a project that’s like, okay, this would be fun to do. Have a problem you’re solving that forces you to actually do this thing.
Howie Liu: For sure. Look, I think that can be night and weekend projects, it can also be the daytime job projects, right? I mean, I am basically telling our teams on the AI platform group especially, “Look, in that low hanging fruit metaphor, it’s like I’m not being prescriptive with you on which watermelons you should pick, but you should go …” We do have different pods within that group, but one of them for instance is what we call the field agents team and they’re responsible for the agents that work within your app. So this is not the agent that builds your app, but these agents that run on a customer’s behalf to do web research on your customers, or they can go and analyze a document and in the future maybe do things like actually generate a prototype of a feature from a PRD or from a feature idea.
I’m telling them, “Look, there’s a almost infinite number of superpowers you can give these field agents. I’m not going to tell you which specifically to do. Now you can ask me to weigh in for sure, but you should go and just experiment and prototype a few different directions we could go.” What if you prototype what it would look like to have a deep research implementation in field agents, so that for any given row of data, let’s say in your case it’s podcast guests, you can just click a button or click a button en masse across every speaker you have lined up to do deep research powered by ChatGPT’s own deep research on each of the speakers and have them all laid out side by side in this table, right? Go prototype that and see how it feels and looks like. So I think some of this stuff can also be in your daytime job, especially if that daytime job is literally to go and build AI functionality.
Lenny Rachitsky: I actually tried to do exactly that. The problem I ran into, I wonder if it’s changed, is there’s no API for ChatGPT deep research yet as far as I know.
Howie Liu: There is now, there is now.
Lenny Rachitsky: There is, there we go.
Howie Liu: Sometimes it ends up being … and I think they only recently exposed it. It ends up being something on the order of a dollar plus per research call, which-
Lenny Rachitsky: What a deal.
Howie Liu: … I mean, again, exactly. I mean, some people would say, oh my god, that’s so expensive, and you rack up 50 of those, you’ve cost $50 a month. I think it’s like, well, it just saved you hours of research by a human.
Lenny Rachitsky: Not only that, I actually have a researcher that I pay to give me background on guests that was four or 500 bucks and the dollar sounds great. I’ve been doing this-
Howie Liu: [inaudible 01:03:28]
Lenny Rachitsky: … I’ve been doing this manually.
Howie Liu: If he was being smart he would be using deep research and they just collected [inaudible 01:03:33]
Lenny Rachitsky: They might be. They might just be. Oh, man. Okay, there’s one more skill I wanted to talk about real quick. This comes up a lot in these conversations is evals.
Howie Liu: Okay.
Lenny Rachitsky: The power of getting good at evals, I know that’s something you value highly. Talk about just why you think this is something people need to get good at.
Howie Liu: Yeah, and I listened to your episodes with [inaudible 01:03:54] and Mike who talked about this. I think it’s interesting that both heads of OpenAI and Anthropic have converged on this point. I mean, look, I think I would add a slightly different or additive take though, which is I think for a completely novel product experience or form factor, you should actually not start with evals and you should start with vibes, right? Meaning you need to go and just test in a much more open-ended way, like, does this even work in kind of a broad sense?
So as an example, for our custom code generation capability, instead of defining evals that get repeatably tested as you vary the prompt or the model or the agentic workflow used to generate these outputs, and you have to define what does good look like by definition for the eval, I would first start with a much more open-ended and ad hoc style of just throw stuff against the wall, try different prompts and see how well it does.
To me, evals are more useful, A, once you’ve converged on the basic scaffold of the form factor and you kind of know what are the use cases you want it to work well for and what you want to test against it. Whereas in the early days, especially if your product market fit finding either for an entirely new company or for a pretty dramatically new or bold new capability that doesn’t really have … it’s not an incremental improvement on something that exists in Airtable today, I think you have to just be a little bit more creative initially and throwing stuff at it, seeing what works to understand, okay, let’s use an example, we’re implementing this new capability that can use basically a long-running AI crawler agent that goes and researches the web for a specific type of object or entity, right?
So it’s similar to deep research, but what it actually does is instead of outputting a report, it’s actually going and compiling a list of things. The things could be companies or people or anything else, right? Find me every Marvel movie ever made, find me every DC Comics spin-off series, literally anything. You have to go in and first just try out a bunch of random … use your own brain to think of what’s the range of use cases I can test this against, right? Then you get back some results and you’re like, okay, well, it’s clear that where it does really well are these types of searches, people and companies with this kind of parameter.
I think to me, evals are useful once you have a sense of what is that cluster of useful use cases, you can start then more programmatically measuring the changes that you’re making to improve the output for that, right? But by that point, you’ve probably already scoped the product and maybe the way we would merchandise it in Airtable is not a completely open-ended capability, but hey, here’s a specific capability that can research one of these X number of entity types including people and companies, and here’s even the filter conditions or criteria that are more explicit that you can define to give it the prompting to search for that thing, right?
But I kind of think it’s more useful as a way to iterate your way to improvement, and you can start really testing stuff empirically, right? You can A/B test, especially if you have the scale of a really large product like Anthropic or OpenAI, you can just test everything and see like, oh, this model actually performs better than this one, this prompt performs better than this one, but I think early on you don’t have that luxury and you’re in a much more open-ended discovery process.
Lenny Rachitsky: That is very wise, evals could constrain you too early. I think about just the Double Diamond, I don’t know, IDO framework of be divergent first, and then converge and then maybe-
Howie Liu: Yeah. Yeah, exactly. I hadn’t heard that before, but that completely resonates.
Lenny Rachitsky: Okay, let me try to reflect back some of the advice I’ve been hearing about how to shift a company to be successful in this new world, and let me see if I’m missing anything that you think is really important. So, one is there’s this sense of just reset the expectations on pace and urgency and help people understand in AI things move incredibly fast, this is how we need to operate. Then there’s also a piece of get stuff out so that you can learn how people use it and what it’s capable of versus polishing it endlessly. Forcing people almost … I don’t know if forcing’s the right word, but encouraging people to play with the latest stuff and giving them a chance to take days off or block out calendars, cancel meetings, just stay on top of this stuff to play as you talked about it. Then sharing things they’ve learned, get the vibes of what’s possible.
Howie Liu: Yeah.
Lenny Rachitsky: There’s also this idea of just rethink, okay, if we were just start today in this world, what would we do to achieve the same mission we are trying to achieve? Ideally it leverages this unfair advantage we have with things we’ve been working on for a long time. Then there’s just talk to AI constantly every hour as you described.
Howie Liu: For sure. Yeah, multiple times an hour, if possible.
Lenny Rachitsky: Multiple times an hour, it keeps going up. Is there anything else that I missed there that you’re like, you need to do this too to have a chance?
Howie Liu: I think just to really, really try to break down role silos, and I think that’s true certainly for EPND in the typical EPD triangle, but I also think it’s probably true even for non-product roles, right? I think it’s true in marketing, right? Something I’m really pushing for in marketing and I think our marketing team is really leaning into actually is if you can just do all of the thing yourself … traditionally how a marketing team might operate is like, okay, you have one person who’s responsible for executing the performance marketing part of a campaign. They literally go into the Google AdWords interface and they’re tweaking the parameters of targeting and budget and conversion tracking, et cetera, and then somebody else is actually responsible for coming up with the specific ad copy, and somebody else yet was responsible for coming up with the seed content or positioning guide written by a PMM that feeds into the ad creative, and so on and so forth, right? Maybe they’re promoting some new demo asset that somebody else yet created.
I just think that in the same way that you can collapse the roles in EPD, and the ideal person, maybe they’re very specialized and deep in one dimension like engineering, but they’re well-rounded enough to be dangerous on the other two, I think that’s kind of true in almost every other function, right? Sales as well, I think you should start to be able to play more of an SE role. Traditionally salespeople didn’t necessarily know the product that well and relied on the SE to come in and be the product experts. I think it’s really hard to sell any kind of AI product now without actually being fluent in the product and be able to demo the product, so AEs need to be SE fluent as well.
So I just think that that concept of collapsing roles, everybody needs to become more full stack to do the … being more outcome-oriented, right? Your outcome as an AE is to convince customers of the value of your product and close deals, right? Okay, well, in order to do that, you used to have dependencies on having assets created by marketing and an SE to help you demo. Can you collapse more of those dependencies so that if you had to, you could do it all yourself, right? I just think it’s a new operating mentality overall for every AI native company or company that wants to compete in this new arena.
Lenny Rachitsky: That is a great addition. It almost feels like you go back to startup times when everyone’s doing a bunch of stuff. There’s no here’s the head of product, here’s the head of engineering, we’re just doing stuff-
Howie Liu: Totally.
Lenny Rachitsky: … that needs to be done.
Howie Liu: Totally.
Lenny Rachitsky: Yeah, I’m kind of seeing it as this upside down T where there’s the thing you’re really strong at and then as you described, the minimum of being good at engineering design or … and SE, by the way, sales engineering imagine is what that stands for. Adjacent roles, you need to start having a baseline. The baseline is increasing of how much you need to understand that, everyone’s Venn diagrams are kind of converging.
Howie Liu: Exactly.
Lenny Rachitsky: Amazing. Okay, let me take a step back and zoom out and think about the broader journey you’ve been on over the past decade plus. Let me just ask you this, what’s the most counterintuitive lesson you’ve learned about Airtable building and company building teams that maybe goes against common startup wisdom?
Howie Liu: I heard your interview with Brian Chesky and then later you talked about founder mode in that YC retreat, and the points there really, really resonated with me. I feel like maybe less eloquently I deduced some of the same principles just in my own experience, which is I think when you’re scaling up, and this relates also to what we talked about before around the early days of building a company, you’re in the details, you’re finding product market fit, you kind of have to be pretty versatile, right? All these decisions from a technical standpoint to design, to even commercial, and what’s the freemium model going to be like? And how are we going to market this product? What does the website look like? They’re all very intertwined, right? You can’t compartmentalize and then almost factory produce each of these things separately. They’re all intertwined and you have a very small tight-knit team that’s thinking full stack about all of this combined.
Obviously that’s the only way, in my opinion, to create that magical product market fit in the first place. Then I think as you scale up, the default guidance that you often get from operational experts and larger scale company investors is like, okay, you got to industrialize the process of all of this stuff, right? It’s kind of like going from a bespoke artisanal, one person made an entire item of clothing to we got to factory produce this thing, right?
What that means in an organizational context is you then create these different fiefdoms, you hire all these execs and each exec just manages their own swim lane, and there’s relatively looser coupling between all of those different groups, right? So then you’ve got sales executing on its own thing, marketing’s executing on its own thing, product’s executing on its own thing. Even within product there’s different product groups and surface areas that are each executing on their own thing.
Using the factory metaphor of there’s an argument that that’s actually kind of an efficient way to scale up production for each of these different swim lanes, right? Each one can operate in a more autonomous and purely scale up focus, wait, how do we produce more of this thing? If the thing happens to be within one product group improving search, that’s our main focus. We’re just going to go and ship, ship, ship more stuff to improve search. So it’s not completely crazy why people give this advice, but I think what you lose is the magical integrative value of holistic thinking and making the bigger picture bets, right?
I think Brian talked a lot about this on his episode with you, which is like, look, in a company that is really serious about product, first of all, I really liked his point about the CEO has to play a CPO role, you have to care about the product. Ultimately the product is the thing and you can’t just coast on scaling up go-to-market around the product forever, you got to keep innovating on the product. By the way, the best way to innovate on the product is not incrementally split over all these different little surface areas, but actually to have a bigger, more step function vision of how this product needs to make a leap, or what’s the next big either act of the product or new capability of the product or reinvention of the product, right?
So I think if you really care about doing that from a product execution standpoint and almost refinding new product market fit on a regular basis, I think it necessitates a completely different operating and leadership model throughout the organization. All of the stuff we just talked about in terms of how to operate in the AI native era I think is actually exactly the same as how you need to operate in this constant product market refinding of fit state.
So I could not agree more with that concept of you got to think ambitiously and move the organization holistically towards these bigger outcomes, but also ship and learn and experiment a lot more in this era. Then maybe the meta learning I had from all of the above is that the specific advice obviously was like, okay, go scale up in this way or go hire these types of people, experienced operators, et cetera. Now, obviously there’s some truth to that, right? The people giving this advice are not incompetent. They had some reason for giving it and in certain contexts that is the right thing to do, but I think my meta learning is it’s not enough to just trust the recommendation, like, here’s the action you should take from a lot of people, ‘cause everybody has different priors and it’s almost like we’re all our own LLMs, and we all have different training from a different corpus of data informed by our own experiences. Maybe you’re trained on the service-
Howie Liu: … experiences, and maybe you’re trained on like the kind of ServiceNow or the Oracle training corpus, and this person’s trained on the Facebook corpus, and I’m trained on the Airtable one. I think what I’ve tried to do more and more is not to just ignore advice from smart people. Obviously, that’s not the right answer, but to kind of take their… It’s almost like in an LLM you can now with a reasoning model actually inspect the chain of thought and see how it’s thinking. Why did it come up with this answer? To me, that chain of thought like “Why did you recommend this?”, is actually more informative than the actual, “Just do this recommendation.”
The answer might be like, “Hey, at So-and-So company, this is how we eliminated the PM role entirely.” For Brian at Airbnb, it made sense. We’re no longer having PMs in their traditional form. Now, we have program managers and product marketers, but more than the actual decision because I don’t think it’s a one-size-fits-all, everybody should do the same, why did you do that? The why actually was very informative, and then be able to take that and say like, “Okay, how would I apply that?” Maybe it yields a different outcome, but the reasoning actually is very informative.
Lenny Rachitsky: It’s interesting how this idea founder mode is not so different from this ICCO trend that you’re following and it’s-
Howie Liu: For sure.
Lenny Rachitsky: … yeah, yeah, it’s like being in the weeds, being in the details, trying things yourself, not delegating to execs.
Howie Liu: Yeah, and I think anything taken to an extreme can be problematic. There is a world where you are so in the details and in every detail that you’re basically just micromanaging and you’re kind of creating like a euphemism for that. That’s not really what founder mode is about. That’s not like the Brian conception of founder mode is to like micromanage everything and not trust anyone, but I think it’s more about finding that right balance of being unabashed about caring about the details that do matter and where the tying together of details across different groups or departments actually is the only way to yield a non-incremental outcome. Otherwise, each person is just optimizing within their own domain, but you’ll never get to the global maxima or the global breakthrough.
I think the really cool thing about CEOs as I seize it, frankly any leader playing more of an IC-like role and being in the details is I think for the right type of person, it’s actually more fun that way. I mean, to be honest, for me, the times where I felt most disintermediated from what I felt was the substance of this company was when I thought that I was almost like forcing myself to step away from the details. I thought that’s what a at-scale CEO was supposed to do. I mean, there’s some famous CEOs who have talked about, “The less decision I could make the better. The less details I’m exposed to the better. I just want to inspect at the topmost layer how this business is running, and if everything underneath it is going smoothly, then I’m able to do that and everything looks good.”
I just think that’s maybe, again, it works in a certain type of very mature type of business. Even then, though, I can’t imagine that at a CPG company like a Procter & Gamble. You wouldn’t want to have a CEO who still actually goes and tastes the soup and tries the products and sees literally the details of what the new product innovation pipeline looks like, as well as like how it’s being experienced on the shelves and so on. I don’t know. I guess I’m just more and more skeptical that that hands-off pure delegation and process management role ever works as a CEO. Maybe you go through a long enough period of where the business is coasting that nobody notices, but I got to say, for me it’s just much more invigorating to get to play that role. I think for the types of operators and leaders that I most admire, that’s what makes the job interesting. They don’t want to have a automated away kind of role as a leader.
Lenny Rachitsky: If you could go back in time and whisper something in a decade-ago Howie’s ear that would have saved you a lot of pain and suffering over the last decade, what would that be?
Howie Liu: Don’t step away from the details that both you love. I mean, first of all, if your passion is building product and product design, even if it feels like at times the company needs to do all this other stuff like scale up, go to market, and operations and just have like a large people organization, that itself creates a lot of need to do things and manage. There becomes a new job invented just to manage a larger group of people, and obviously you’re going to have to do some of that. You can’t just completely eschew all your responsibility as an at-scale CEO, but don’t lose the essence of the thing that you love doing and that really made this product happen and gives this company as many companies that were founded on a magical product market fit finding insight. Don’t step too far away from that, and always make sure that is still your number one, even if other stuff has to also add to your plate.
Lenny Rachitsky: I think people don’t talk enough about this how if someone starts a company that’s an idea they have they’re excited about, it takes off and then you’re stuck on that for a long time, and then even if things are pushed in a direction you’re not as excited about. This point about just remembering what you actually love about it and coming back to that is so important because that’s the only way to keep doing this for a long time.
Howie Liu: I think that’s so true, and to me that’s why there’s always been a difference between entrepreneurs who love the act of building a product or the business, too, versus those who saw a just purely business or financial opportunity that they felt like they couldn’t pass up exploiting or going after. Look, no knock on people who are more the latter, and there’s entire industries where it’s all just about alpha generation. You can go into the private equity business and so on, and it’s just purely it’s rationally about how do I find the alpha? I think that some of the best companies, product central companies, at least in my opinion, are run by those people who actually just love the product. I think you get a feel for that from some of the AI companies like Sam, I think genuinely just loves working on AI.
If he could spend a hundred percent of his time on just being close to the AI and the research, I mean, he would and he’s even said as much. Ranging to like Brian’s with Airbnb, it’s pretty clear that people like this are not motivated like… Airbnb was not founded because like, “Oh my God, we want to make a lot of money off this arbitrage opportunity against hotels.”
Lenny Rachitsky: They just needed to pay their rent.
Howie Liu: Yeah. Well, that and I think they loved the product and I think they also loved the way in which they built the product, the design-centric nature of that product and company and culture. That’s what gives you the continued joy of working on what could be the same company for a very long time.
Lenny Rachitsky: Howie, is there anything else that you wanted to touch on or leave listeners with before we get to our very exciting lightning round?
Howie Liu: I just want to reiterate, especially for listeners here who are in an EP or D role and especially in the P role, I really do believe that this is not like you either have or you don’t like in terms of the skill set needed to be relevant and AI needed, but I do think it’s a call to action to go and bolster your skill sets where they may be less refined right now. I think even programming, I really believe everyone could learn how to be a software engineer if they wanted to. Now, obviously, some people just as with like great writers are never going to be a published author or the Hemingway, but everyone can gain a good enough proficiency of software engineering if they really wanted to.
You could take that boot camp. You could do like some coding exercises on the side, et cetera. The point there is that sometimes I think we treat these disciplines like hard, hard skills that if you’re already halfway into your career and you’re not already an engineer, if you’re not already a designer, okay, well, you can never be one. I just think our brains are malleable and there’s a lot of great curriculum out there to learn. Lot of it, like I said, just comes down to also like trial and error and building projects, maybe nights and weekends projects even to learn this stuff. Everyone can learn how to be a versatile kind of unicorn product engineer/designer hybrid in the AI-native era. The only thing stopping you is just going out and doing it.
Lenny Rachitsky: That is a really empowering way to end it, and just to double down on that, it’s never been easier to learn these things. There are super intelligences that you can talk to that do a lot as they’re building can help you learn.
Howie Liu: Yeah. I mean, literally, I go into ChatGPT sometimes and I ask it just like, “Hey, how would you build this app?” I’m just curious. I’m like, “How would you build Manus, the open-ended agent?” Literally, how would you build it? You can ask the questions and it’s like having an amazing, brilliant software architect, software engineer, product manager, designer expert tutor that you can literally like there’s no dumb question. They have infinite patience. They’re literally on and awake 24/7. It is the most incredible time to learn this stuff, to your point. Then, of course, the interactive tools to go and actually build stuff. Anyone can download Cursor and just start asking Composer to generate some code for you, and then looking at the code and trying to figure out what it does. To your point, when I think back to the earliest era that I experienced of building apps, first I learned C++, then I learned PHP and JavaScript and even building kind of JavaScript single-page apps in the early days like ‘08 through 2010. It was a dark, dark art. I mean, there were some like… You just had to go and like learn some of these things. There wasn’t great tutorials for it. You had to reverse engineer certain things. There were just weird things like if you wanted rounded corners in your UI, you literally took Photoshop, opened it up, created like a rounded corner in pixels, and then cut that pixel up into an image that you dropped onto the page at exactly the right position to be at the edge of a box.
It’s like crazy stuff. I mean, everything was so much more arcane at the time, and now it feels so much more fluid and accessible, and the gap between the arcane tech that you have to wade through to build something has just been minimized so much. It’s like the effort and abstraction between you and the magical, delightful actual building of the thing that you want has been so minimized. It’s never been a more exciting time to be a builder.
Lenny Rachitsky: You remember spacer.gif?
Howie Liu: Oh yeah, yeah.
Lenny Rachitsky: It’s like to create. It’s that line stuff you just kind of have-
Howie Liu: Yeah, I remember it. Yeah.
Lenny Rachitsky: … the invisible one-pixel thing that you just stick in places.
Howie Liu: Yeah. Yeah, yeah. No.
Lenny Rachitsky: Oh my God, what a time to be alive. Howie, with that, we’ve reached our very exciting lightning round. I’ve got five questions for you. Are you ready?
Howie Liu: Yes.
Lenny Rachitsky: Here we go. What are two or three books you find yourself recommending most to other people?
Howie Liu: You know, I’ve been trying to read fiction more, partly because I think it’s just a really nice mental reset. I will say like Three-Body Problem for anyone who hasn’t read it, it’s a mind-expanding book. I like sci-fi and fiction that kind of opens your brain, so maybe this is my cheat card, but it’s a three-book series. Those are three great books.
Lenny Rachitsky: I love that series, and my tip there is it gets good one and a half books in is my tip, so just keep reading. That’s where it’s like, “Okay, now I’m in.”
Howie Liu: I liked even the first one, but I felt like it was inception where every subsequent book was like you dropped into another, like you incepted into another layer, right?
Lenny Rachitsky: Awesome. Okay. What’s a favorite recent movie or TV show you’ve really enjoyed?
Howie Liu: TV show, I just started watching The Studio. It’s like the Seth Rogen, Rogen.
Lenny Rachitsky: Yeah, it’s so stressful.
Howie Liu: Yep. Yeah, it is pretty stressful, and I mean, Silicon Valley was too close to home when it came out, so I watched it, but it was just cringy. The Studio is kind of fund to watch because it’s a little bit about like inside baseball of Hollywood, and yet I’m not in Hollywood, so it’s entertaining to watch. It’s I thought smart and a funny show because I split time between L.A. and S.F. I also feel like it’s very real to me. I see a lot of the literal characters out there in the world that it’s characterizing.
Lenny Rachitsky: Do you have a favorite product you recently discovered that you really love? Could be an app, could be gadget, could be clothing.
Howie Liu: Okay, so I’ll give two because I feel like I have to say some kind of software product. I mean, I’m a really big fan of Runway, the product and the company. I just think every new model they come out with, they just came out with a new one just I think like two days ago that gives even more controls and refinement on creating exactly the video scene that you want. I think just the photorealism in what you can generate now, and they also built this cool demo thing that’s an immersive world generator I mentioned before. I think it’s just cool to see. I also like the underdog story. I’m clearly like Google’s gunning in the space, has VO3 and so on and has its OpenAI, but I love the underdog story of this sub-hundred-person company still punching above their weight and building really awesome video experiences. That’s the software one.
Then, a very, very kind of nerdy real-world answer on product is I kind of just recently got into this whole cottage industry of artisanally produced basically clothing by small-scale Japanese manufacturers that use literally like hundred-year-old looms to make clothes the old-fashioned way or the old-fashioned industrial way. They have these loop wheeler machines and they spin the cloth in a very slow pace, so it’s completely impractical from a production-scale standpoint, but I’ve gotten some of these t-shirts and I just love the… I guess in a world where it feels like everything is becoming so much faster moving and even tech from five years ago is obsolete, I love a little bit of the throwback to like old things sometimes can be even more cherishable in this new era. Maybe that makes me a hipster, but I love the vintage, the retro increasingly these days.
Lenny Rachitsky: I feel like anything that starts with artisanal small batch Japanese is going to be really good stuff. Is there a brand you want to share that is that? Or is this like you want to keep it-
Howie Liu: Yeah. No, I mean-
Lenny Rachitsky: … under the radar.
Howie Liu: … actually, so Self Edge, which actually has a storefront, the main storefront is on Valencia Street in S.F. They carry a lot of these items. That’s kind of their whole MO and they have like jeans and t-shirts. I’ve gotten a lot. I mean, they basically curate a really good selection of different actual makers. One of them is called Studio D’Artisan, another one’s called… Actually, it’s cool. There’s this company called… I think the umbrella company is actually just Toyo, T-O-Y-O, Manufacturing, which sounds like it’s a big kind of like large-scale conglomerate, but it’s anything but. It’s like a really small-scale Japanese vintage manufacturer of clothing, but they have a few sub-brands.
They actually bought the rights to this American post-war brand that was kind of like Hanes, one of the like big four or five menswear, kind of undershirts and athletic wear brands called Whitesville. I don’t know where the name came from, but basically it’s a bunch of like basic clothing, like t-shirts, et cetera, and this Japanese indie company, they bought the defunct basically name and now is reproducing clothes almost made to the exact shape and stack, and even with the exact recreation of the graphic packaging on these tees, but like today. I just think there’s something really funny and ironic about they’ve taken an American post-war aesthetic and literal brand, but it’s actually a indie small-scale Japanese manufacturing approach to making those clothes.
Lenny Rachitsky: I feel like we just tapped into what could be a whole other podcast conversation about clothing and-
Howie Liu: Yeah-
Lenny Rachitsky: … craftsmanship-
Howie Liu: … [inaudible 01:35:57].
Lenny Rachitsky: … but I’m going to pull us out of that.
Howie Liu: The next podcast franchise.
Lenny Rachitsky: Or just Howie and Lenny talking about clothing.
Howie Liu: That’s great.
Lenny Rachitsky: Okay, two more questions.
Howie Liu: Yeah.
Lenny Rachitsky: Do you have a life motto that you often find useful in working or you like to share with friends or family?
Howie Liu: I stumbled on this guy Paul Conti, who I think he’s an MD, but also a psychologist, and he has a book, but also he did this long-form podcast with Andrew Huberman, and he actually ends up talking a lot about just how to think about your life outlook and kind of your framework for thinking about life, but grounded in a kind of like scientific and neurological and cognitive science basis. I found one particular point really, really powerful it took with me, which is if you live your life in a way that’s foundationally built around humility and gratitude. Look, everybody has different circumstances.
I think I fully own that even though I didn’t come from money, my family was very, very financially modest growing up. I still had incredible resources and opportunities afforded to me even just by virtue of growing up in the U.S., being born in and growing up in the U.S., but also having access to a computer and the internet and even all the free resources I could then access and learn about from there. I still feel like whatever you have or don’t have to start with, if you kind of approach the world and kind of the future with a spirit of humility and gratitude rather than, I guess, the opposite of that, I think I’ve felt like it kind of becomes a self-fulfilling prophecy. You’re open-minded, you’re kind of grateful, and then more opportunities actually come your way, and maybe it’s because the energy you’re putting out into the world and other people.
You’re kind of attracting good opportunities and good people and good things, but I think there’s a lot of other parts of his framework, but the one that is easiest to remember is like, how do I approach each day? Even if I’m going through a tough moment and I had to fire somebody today, or maybe I get disappointed because we lost a customer deal or something broke or whatever, but to still try to look at the entire situation from overall a feeling of humility and gratitude I think just really does shift your like… It spills over into everything else for that day and maybe even for the whole lifetime.
Lenny Rachitsky: That super resonates. That is really powerful advice that’s hard to internalize, but important.
Howie Liu: Yeah, it’s easily said, hard to practice.
Lenny Rachitsky: Yeah. Where can folks find you? What should they know about Airtable and how can listeners be useful to you?
Howie Liu: Okay, so I am on Twitter, howietl. I don’t post that much, but I’m a lurker, so I listen and watch, and you can always DM me there. You can also email me directly, howie@airtable.com, anytime you have ideas, feedback, et cetera. On Airtable, just go try it. The whole point is we want to make this an experiential product. That’s why we’re really leaning into the PLG roots. We talked about the homepage literally says like, “Just start building right now. What do you want to build? Go.”
It starts building, and so use the product, give me feedback, and if you have ideas of your own and you want to rip on them, I love because my passion is thinking about product and product UX, especially in the AI era if you’re working on or thinking about something interesting in that space. Even if it’s just purely to riff on a concept, that’s something I enjoy doing, and maybe I get to learn and sharpen my own skill set from. Feel free to reach out and, yeah, I mean, tell your friends and family to try Airtable as well. That’s the main thing.
Lenny Rachitsky: Sounds like you’re looking for people to nerd snipe you and-
Howie Liu: Yes. Yeah.
Lenny Rachitsky: … Howie, thank you so much for being here.
Howie Liu: Awesome. Thank you, Lenny.
Lenny Rachitsky: Bye, everyone. Thank you so much for listening. If you found this valuable, you can subscribe to the show on Apple Podcasts, Spotify, or your favorite podcast app. Also, please consider giving us a rating or leaving a review as that really helps other listeners find the podcast. You can find all past episodes or learn more about the show at lennyspodcast.com. See you in the next episode.
Glossary
| English | 中文 |
|---|---|
| A/B test | A/B 测试(对照实验方法) |
| AE | AE(Account Executive,客户经理) |
| affordances | 可供性(交互设计术语,指界面元素暗示的操作可能性) |
| agent | agent(智能代理,保留原文) |
| Airtable | Airtable(产品名,保留原文) |
| All In | All In(播客名,保留原文) |
| alpha | alpha(指具有前瞻价值的信息/洞察,保留原文) |
| Andrew Huberman | Andrew Huberman(人名,保留原文) |
| APM | APM(Associate Product Manager,产品经理培训生项目,保留原文) |
| Brian Chesky | Brian Chesky(Airbnb 联合创始人兼 CEO,人名保留原文) |
| CB Insights | CB Insights(公司名,保留原文) |
| ChatGPT | ChatGPT(产品名,保留原文) |
| chief of staff | 幕僚长 |
| Composer | Composer(Cursor 的功能,保留原文) |
| CRM | CRM(Customer Relationship Management,客户关系管理系统,保留原文) |
| Cursor | Cursor(AI 代码编辑器,保留原文) |
| Dan Shipper | Dan Shipper(人名,保留原文) |
| decacorn | decacorn(百亿美元级独角兽) |
| Double Diamond | Double Diamond(双钻石设计框架,英国设计委员会提出的设计方法论) |
| DRI | DRI(Directly Responsible Individual,直接负责人制度,保留原文) |
| DSL | DSL(Domain-Specific Language,领域特定语言,保留原文) |
| Eames chair | Eames 椅(Charles & Ray Eames 设计的经典座椅) |
| EPD | EPD(Engineering/Product/Design,工程/产品/设计部门,保留原文) |
| evals | evals(评估/评测,AI 模型输出质量的系统化评估方法,保留原文) |
| Every | Every(公司名/品牌名,保留原文) |
| fast thinking / slow thinking | 快思考 / 慢思考(源自 Daniel Kahneman 的《思考,快与慢》) |
| field agents | field agents(Airtable 中在应用内代替客户运行的 agent,保留原文) |
| Flexport | Flexport(公司名,保留原文) |
| founder mode | founder mode(创始人模式,指创始人深度介入运营的管理方式) |
| Framer | Framer(原型设计工具,保留原文) |
| freemium | 免费增值(免费基础版+付费高级版的商业模式) |
| GitHub Copilot | GitHub Copilot(产品名,保留原文) |
| GM | GM(General Manager,总经理制,保留原文) |
| golden pathy | 理想路径的(指测试中的最佳场景路径) |
| GPT-5 | GPT-5(OpenAI 模型,保留原文) |
| growth mindset | 成长型思维 |
| Harvey | Harvey(AI 法律科技公司,保留原文) |
| HeyGen | HeyGen(AI 虚拟人视频生成平台,保留原文) |
| hipster | hipster(指追求小众品味的人,保留原文) |
| HyperDB | HyperDB(Airtable 的数据存储引擎,保留原文) |
| IC CEO | IC CEO(个人贡献者型 CEO) |
| jobs to be done | jobs to be done(待完成的任务理论,保留原文) |
| loop wheeler | loop wheeler(一种低速圆筒编织机,保留原文) |
| Lovable | Lovable(AI 应用构建平台,保留原文) |
| map reduce | map reduce(分布式计算范式,保留原文) |
| nerd snipe | 用有趣的问题吸引某人的注意力(网络用语,指用一个引人入胜的技术/智识问题让人无法抗拒地去思考) |
| Nick Turley | Nick Turley(人名,保留原文) |
| Nikita | Nikita(人名,指 Nikita Bier,保留原文) |
| no code | 无代码 |
| NotebookLM | NotebookLM(Google 的 AI 笔记工具,保留原文) |
| Omni | Omni(Airtable 的对话代理,保留原文) |
| OpenAI | OpenAI(公司名,保留原文) |
| Palantir | Palantir(公司名,保留原文) |
| Paul Conti | Paul Conti(人名,保留原文) |
| PLG | PLG(Product-Led Growth,产品驱动增长,保留原文) |
| PMM | PMM(Product Marketing Manager,产品营销经理) |
| POC | POC(Proof of Concept,概念验证,保留原文) |
| PRD | PRD(Product Requirements Document,产品需求文档,保留原文) |
| product-market fit | product-market fit(产品市场契合度,首次出现保留原文) |
| Replit | Replit(在线编程平台,保留原文) |
| Revolut | Revolut(数字银行/金融科技公司,保留原文) |
| Rippling | Rippling(公司名,保留原文) |
| Runway | Runway(AI 视频生成平台,保留原文) |
| SE | SE(Sales Engineer,销售工程师) |
| Self Edge | Self Edge(品牌名,保留原文) |
| Sesame AI | Sesame AI(AI 语音技术公司,保留原文) |
| Seth Rogen | Seth Rogen(人名,保留原文) |
| Studio D’Artisan | Studio D’Artisan(品牌名,保留原文) |
| v0 | v0(Vercel 的 AI 代码生成产品,保留原文) |
| vibe coding | vibe coding(凭感觉用 AI 写代码的方式,保留原文) |
| vibes | vibes(凭直觉感受,与 evals 的系统性测试相对,保留原文) |
| Whitesville | Whitesville(品牌名,保留原文) |
| Windsurf | Windsurf(AI 代码编辑器,保留原文) |
| YC | YC(Y Combinator,知名创业孵化器,保留原文) |
Reformatted by reformat_english.py
我们如何为 AI 重构了 Airtable 的整个组织架构 | Howie Liu(联合创始人兼 CEO)
访谈实录
Howie Liu: 如果你真的是带着同样的使命从零开始创立一家新公司,你会如何用完全 AI 原生的方式来执行这个使命?如果你做不到,那你应该找一个买家;如果你真的在乎这个使命,那就去开始它的下一次轮回。
Lenny Rachitsky: 那么为你工作的人呢,你是如何调整对他们的期望,来帮助他们取得成功的?
Howie Liu: 如果你想取消一整天甚至整整一周的所有会议,去把每一个你觉得可能与 Airtable 相关的 AI 产品都玩一遍,那就去做吧。
Lenny Rachitsky: 在产品团队的不同职能中——PM、工程、设计——谁在使用这些工具提升生产力方面成效最好?
Howie Liu: 这其实更多取决于个人的态度。这三个角色中,任何一个能够跨界融合另外两个角色的人都会有很强的优势。作为 PM,你需要开始更像一个混合型的 PM 原型师,具备一定的设计审美。
Lenny Rachitsky: 你觉得这些角色中哪一个面临的危险最大?今天的嘉宾是 Howie Liu。Howie 是 Airtable 的联合创始人兼 CEO。我在这个播客上正在与一系列创始人对话,他们都在这个 AI 时代重新改造自己十多年历史的业务,以帮助大家应对每一家公司和产品当前正在经历的这场生存性转型。Howie 和 Airtable 的经历就是一个令人惊叹的范例,Howie 在这次对话中分享的内容有太多值得学习的地方。
我注意到一个非常有趣的趋势,Howie 就是一个典型的例子——CEO 们几乎重新变成了个人贡献者,亲自写代码、构建产品、主导项目。我们把这种现象称为 IC CEO。我们还会谈到他认为产品经理和产品负责人,以及工程师和设计师需要培养哪些具体技能,才能在我们所处的这个新世界中脱颖而出。另外,他还谈到了如何将公司重组为两组——一个快思考组和一个慢思考组——这让他们的 AI 投资显著加速。如果你正在苦苦思索如何在这个新的 AI 时代取得成功,这期节目就是为你准备的。
(广告段落已跳过)
Lenny Rachitsky: Howie,非常感谢你来参加节目,欢迎来到播客。
Howie Liu: 我非常兴奋。谢谢你,Lenny。我作为远方的听众已经关注了很久了。
Lenny Rachitsky: 听到你这么说我很受宠若惊。我也非常期待这次对话。过去大概——是 13 年吗,还是更久?你经历了一段相当不平凡的旅程。
Howie Liu: 对,差不多 13 年。
“Airtable 已死”的病毒式推文
Lenny Rachitsky: 13 年。我想其中有很多高潮也有很多低谷,这些我都想聊。我想谈谈你一路上学到的很多经验教训。我想从一段我猜测在 Airtable 历史上非常令人意外的低谷时刻开始。不幸的是,当我想到 Airtable 时就会想起这件事。我觉得其他人可能也有同感——有一条推文曾经超级病毒式传播,大概到现在已经有几年了,有人分享了各种数据,然后说,Airtable 完了。他们融了远远超过自身价值的钱,赚的钱根本不够从水下浮上来。是的,Airtable RIP。到底怎么回事?那些说法有多少是真的?后来怎么样了?
Howie Liu: 其实基本上没有什么是真的。让我感到意外的是这条推文传播得如此之广,因为……说实话,我回头看了这个人的其他推文,他好像在 CB Insights 工作,而讽刺的是那家公司的整个卖点就是拥有关于私营公司的优质数据和良好的数据质量。但他们的数字完全就是错的,而且差了很大的倍数——无论是我们的收入规模还是增长率。让我稍感安慰的是,我回看了这个人之前也发过关于其他公司的推文,比如上一条针对 Flexport 的做空推文。他们说什么”Flexport 完了”,估值太高了,等等。所以,我觉得更令人意外的是,这个人一直在发一堆没有任何真实数据或正确数据支撑的刺激性言论,而偏偏这条推文超级病毒式传播了——这才是让我困惑的地方。然后,真正让这件事持续发酵的是 All In 播客,那个播客显然超级受欢迎。我听了那一期,他们讨论了这件事,说”本周新闻,关于 Airtable 的这条推文,大家怎么看?“我觉得这几乎变成了一个谈论更广泛话题的由头——那些上一代高估值公司,也许是 decacorn(百亿美元级独角兽)公司,在新的……在那个时间点,是公开和私有市场最近的低迷期,会面临怎样的命运。不过他们后来确实也发布了更正。All In 在几周后做了一期后续节目,说”嘿,我们的数字搞错了,我们正在修正对 Airtable 的判断和看法。”
Lenny Rachitsky: 那句话怎么说来着——谎言已经绕了地球好几圈,真相还没来得及起床?
Howie Liu: 是的。在那次经历中,我对模因和传播规律学到了很多。我本来不是一个很擅长社交媒体的人,但我觉得自己多少长进了一些。
Lenny Rachitsky: 确实很难。Twitter 就是个……它的激励机制完全错位了。我发的都是人们想转发的东西,而不是真相。
Howie Liu: 嗯,尤其是……我是说,Twitter 也有很多值得肯定的地方……我会说,我更喜欢 Elon 接手后的 Twitter 而不是之前的,因为它更大胆了,而且我真的很欣赏大胆的产品执行——不是躺在过去的功劳簿上,他们做了大量的改动。但我确实觉得,我的信息流里不断被注入耸人听闻的内容,而且,说实话,这对我确实有效。我忍不住会去点击、去互动。但这确实……我觉得它导致了这类内容的疯狂传播。
Lenny Rachitsky: 对。现在 Nikita 在掌舵,不知道你有没有看到,有个新功能……我们不用一直聊 Twitter,但有个新功能是你截一张推文的图,右上角会有一个巨大的 X.com 水印。对,就是……人们一直在到处分享这些推文截图。
Howie Liu: 嗯。
Lenny Rachitsky: 天哪。那边从来不会无聊。
Howie Liu: 确实。
CEO 回到一线写代码
Lenny Rachitsky: 好,我想聊一个完全不同的话题,这也是我非常期待和你讨论的——我注意到一个新兴趋势,而你似乎走在了最前沿:CEO 重新变回 IC。也就是所谓的 IC CEO——CEO 们再次卷起袖子亲自动手,重新写代码,重新深入技术细节。我觉得你同样走在这个趋势的最前沿。聊聊你为什么这么做,为什么觉得这很重要,以及现在每天的工作状态和几年前有什么不同。
Howie Liu: 这一转变背后的原因,至少对我而言,是这样的:公司刚起步的时候,我完全就是这种状态。我实实在在地在写代码,既写后端,思考平台的实时数据架构,也做前端、做用户体验。我认为在那个创始阶段,寻找最初的 product-market fit 时——尤其对于一款纯软件产品来说——我们不是在做一个运营密集型的业务,比如遛狗市场平台那种技术只是附属品的东西。
技术本身就是产品。而且很 Meta 的一点是,Airtable 本身就是让其他人搭建自己应用的平台。所以一切都是围绕技术展开的——那些最核心的设计决策,包括架构层面、前端层面、产品用户体验的每一个选择。这就是产品的价值主张,二者不可分割。你不能说:“好,我研究了 jobs to be done,这是工作流、这是流程,然后找个工程师随手实现一下就行了。”
正是那些微小的决策,以及对浏览器端和实时数据架构最前沿可能性的真正把控,成就了这个产品。我觉得 Figma 也是如此,它和我们的时间线几乎平行。我们差不多同一时期创立,都花了两年半时间亲手打造产品,早期团队亲力亲为,然后才发布。当我回想从那个创始时刻到现在的这段时间——包括现在这个新的生成式 AI 时刻——我认为 SaaS 整体和 Airtable 都经历了一个成熟期。
随着公司规模扩大,你学会了如何搭建团队和组织,你不得不把精力放在那些并非亲密细节的事情上——流程、人员等等——你就自然而然地离那些细节越来越远了。也许对某些业务来说这没问题,因为核心不再是找到让产品具有魔力的 product-market fit 的那些细节,而是把一个已有的可行模式规模化,用一些更粗线条的手段去推进——比如一个更粗粒度的路线图,一个更粗放的市场执行策略。
但无论如何,我认为现在我们进入了一个新的时刻——在我看来,每一个软件产品都必须重新创立一次。因为 AI 是如此巨大的范式转变,它甚至不只是从桌面到移动端、从本地部署到云端那种转变——那更多是一次性的、某种程度上可预测的形式因素变化。而 AI 演变速度之快,每一次进化、每一个新模型的发布、每一种新能力的出现,都意味着需要发明新的形式因素和新的交互模式,才能充分释放这些能力的潜力。
所以要在这个时代保持持续的相关性、不断打磨 product-market fit,我认为你必须沉浸于细节之中。你不能站在一万英尺的高处俯瞰然后说:“好吧,我们往这个问题上多投一些人就行了。“而是要真正理解什么才是正确的产品体验,什么才是支撑它的商业模式,以及其他所有需要配合的环节——从而在我们这个产品领域中充分利用这些新能力。
Lenny Rachitsky: 你在某处用过一种说法,谈到做”首席品味官”(chief taste maker)。
Howie Liu: 对。
Lenny Rachitsky: 要做到这一点,就必须像你刚才描述的那样去做。
Howie Liu: 没错。我的意思是,我觉得现在还有一个问题是,如果你不亲自参与制作,你就很难真正品尝这道汤的味道。也就是说,在 AI 时代,你可以看着最终产品说:“嗯,感觉对了”或”不对”,或者”我们的步子够不够大,有没有把这些新能力好好产品化”。但要真正理解什么是可能的解空间,你必须深入细节。
说白了,你不能只看截图或预录的视频来了解一个新的产品功能。AI 是你必须亲手去玩的东西——最好是既玩那个打包好的应用或解决方案,也直接操作底层的基本组件,通过 API 或聊天界面直接使用模型,把它们推向极限。因为这是你真正理解这些新”食材”的唯一方式。就像一个厨师,你突然获得了一批令人惊叹的新食材,但你得先熟悉它们,才能把它们做成一道新菜。
CEO 是否亲自使用 AI 是成功的关键信号
Lenny Rachitsky: 我们之前请过 Dan Shipper 上播客,他运营着一个名为 Every 的通讯和播客,旗下有一家公司,专门帮助企业更好地采用 AI、提升 AI 能力。我问他,什么信号能说明一家公司能成功拥抱 AI 并获得巨大的生产力提升?他说:CEO 是否每天都在使用 ChatGPT 或 Claude?
Howie Liu: 对。
Lenny Rachitsky: 我觉得你描述的正是这个——不是每天,是每小时。
Howie Liu: 真的是每小时,甚至可以用推理成本来衡量,对吧?就是对应的底层推理计算量。
Lenny Rachitsky: 你用了多少 token?
AI 推理成本的最大用户
Howie Liu: 嗯,我很自豪地说,我相当确定自己仍然是——我最近刚查过——Airtable AI 推理成本最高的用户。不仅是我们公司内部,而且在很长一段时间里,我觉得在全球所有客户的账号中我也是第一。我是极其刻意地”浪费”的。所谓浪费,是指我会做一些可能花费数百美元实际推理成本的事情。比如,对销售电话的完整录音文本做大量 LLM 调用,提取不同类型的洞察——比如这是产品方面的要点,或者这是摘要等等。
我们现在还有一个能力,基本上类似于 LLM 的 map reduce。也就是说,即使因为上下文窗口的限制,你无法把整个语料库塞进一次 LLM 调用里,我们也会把所有内容逐一处理,拆分成小块,对每一块执行一次 LLM 调用,然后再对这些块的聚合结果执行一次汇总调用。非常昂贵,因为你基本上是在对大量数据运行一个非常贵的模型,然后又在那些聚合结果上再跑一遍。但对我来说,在这上面花几百美元,相比获得更好洞察的潜在战略价值来说,根本不值一提。
这就像是一个极其聪明的幕僚长,把过去一年我们所有的销售电话记录都读了一遍,然后给我提供非常敏锐的产品洞察、营销洞察、定位洞察和细分洞察。这是无价的。你可以花几百万美元请咨询公司来获得同等质量的工作成果。所以对我来说,AI 的价值与实际成本之比——当你是贪婪但聪明地去使用它的时候——仍然是一个疯狂的比例。更多的人应该积极地把计算资源投入到这些高价值问题上。
Lenny Rachitsky: 直到有人发推特说你让公司在 AI 计算上花了多少钱,你们要亏本了。
Howie Liu: 开个玩笑。就像我们亲自把公司的现金流拉垮了一样。
重塑时间分配
Lenny Rachitsky: 听到这些的 CEO 们、创始人们大概会想,好吧,我也应该开始这样做。具体是什么样子?我想你肯定还有很多其他事情要做——你怎么改变日常安排来做这件事?
Howie Liu: 是的,所以实际上我默认砍掉了大量的一对一会议。我的意思不是说我不想和人单独交流,而是我发现……增加更多固定的一对一会议反而妨碍了我处理更有时效性的议题。我认为最好的会议类型是紧迫性驱动的。就是某个有时效性的话题,你发现了某个洞察。也许我跟某个新创业公司聊过,从他们的产品或方法中学到了一些东西。
然后我想把它带入我们在 Airtable 规划新功能时的思考中,甚至只是给 Airtable 内部一些 EPD 同事种下一颗种子。我希望大多数会议都是很有时效性的,而且由真正的 alpha 信息来驱动的——必须有一定的价值和洞察作为种子。除此之外,我会补充安排一些:当我和某人面对面的时候,我会腾出时间好好叙叙旧,不那么结构化,不那么追求时效性,更多的是建立人与人之间的关系。
但实际上我发现,采用一种……几乎是一种哑铃策略:如果你要以自由的方式和某人共度时间,那就做得高质量一些,而不是强制的每周例行仪式。在条件允许时,一起去吃顿较长时间的午餐,或者边喝咖啡边散步,面对面交流。也许是每一两个月一次。而中间的那些交流则是围绕特定主题的——我们确实有一些固定的会议,比如现在我们每周基本上有一个关于所有 AI 执行工作的冲刺汇报,现在公司一半的 EPD 组织都在做 AI 相关的能力开发。
我们试图非常快速地交付。我基本上一直想问的问题是:一家 AI 原生公司——比如 Cursor 或 Windsurf 等等——他们会怎么执行?我们是否和它们一样快?是否和它们一样充分地利用所有新东西?把这种强度和紧迫感带入我如何分配时间的方式,这是对我来说最大的转变。
组织架构重组
Lenny Rachitsky: 你做了什么改变来帮助公司更快地行动,赶上那种节奏?
Howie Liu: 是的。我们对 EPD 组织做了一次重组。在之前……过去大约四年里我们经历了几次不同的重组。最初的形态,随着组织不断扩张——基本上是默认的、渐进式的——我们有一堆各自负责一个功能或一个产品领域的团队。比如有一个团队负责表格内的搜索,有一个团队负责移动端体验,以此类推。这有它的好处。显然,那个团队可以深入了解代码库的那部分和产品的那部分。
但它的缺点是,是的,当每个人的职责范围实际上就是一个需要他们逐步改进的功能时,你倾向于渐进式地思考。而不是围绕一个使命或一个成果目标来思考——后者可能需要跨越更广泛的多个产品领域来协调重大变革,而不是每个领域各干各地渐进优化。所以,我们最初重组为基本上不同的业务单元。我知道 Airbnb 也做过从职能制到 GM 制再调回来的那种转变。我们这次更像是说,“我们有企业级业务”,他们更多关注的是可扩展性。
能否支持更大规模的数据集和用例?是否具备将一个应用推到一万或两万个席位用于产品运营所需的核心能力?大量的架构工作、大量的扩展性工作。然后我们有所谓的”团队”支柱,更多面向自助服务,偏向产品 UX——采用、上手、分享等基础功能的易用性。还有一个 AI 支柱、解决方案支柱,以及基础设施。
但我们发现这种方案下,虽然确实有了更多整体性的押注——比如团队支柱可以不只是考虑一个功能,而是整体的上手体验,真正以跨越产品多个部分的方式来思考——但还是觉得不够。尤其是当我们开始在 AI 方面加大执行力度时,它无法让我们像一家 AI 原生公司那样快速、激进地行动。你看看 Cursor 这样的公司,它们每周都在发布重大的新东西。而不是”哦,我们有一条企业级路线图,还有另一条路线图”。
它感觉就像一个浑然一体的产品,以极快的节奏在交付。所以我们做了最近这次重组,现在有了我称之为”快思考组”的团队,官方名称叫 AI 平台,但它真正的含义是我们想以接近每周的频率交付一批新能力。而且每一个都应该具有真正令人惊叹的价值——使用 Airtable 的新能力时,你应该惊叹不已。然后另外,我们有”慢思考组”。这不是说哪个更好或更差。字面意思就是,就像人类需要快思考和慢思考一样。
Lenny Rachitsky: 我身后就有那本书。
Howie Liu: 对,我很喜欢那本书。慢思考就是一种不同的规划与执行模式,对吧?它更加审慎,需要更多的预先谋划。你不能指望一周内用个粗糙的原型就上线一个涉及大量数据复杂性的基础设施——比如我们的数据存储引擎 HyperDB,现在能处理数亿条记录的数据集。这种东西不是一周能交付的。所以现在公司有了这两个不同部分,而我觉得真正酷的地方在于,它们实际上配合得非常好。
因为快速执行的部分——也就是 AI 相关的东西——创造了漏斗顶部的兴奋感,激发了新的用例和新用户来到 Airtable,包括大型企业客户。企业也在用这些东西,它不只是中小企业级别的工具。而慢思考部分,则让那些最初的采纳种子能够生根发芽,成长为更大规模的部署。相比之下,我看到很多 AI 原生公司面临的挑战是,它们的漏斗顶部可以很宽——吸引大量 AI 好奇型流量。
大量关注、大量早期使用,但接下来的挑战是,如何将这些转化为更持久的增长,如何让每一颗采纳的种子留下来并随时间不断扩展。
Lenny Rachitsky: 这太酷了。我从来没听说过用这种方式来组织团队——快思考、慢思考,Kahneman 的那个框架。很有意思。对于快思考团队,你有没有发现哪种特定类型的人在那里更容易成功?是不是大量引入新人——那些不习惯 Airtable 既有的工作方式的人?你观察到什么?
Howie Liu: 我们是混合搭配的。我们引入了……我的意思是,我们一直在招人,对吧?公司历史上从来没有停止过招聘。坦白说,即使在我们不得不做两次裁员、大幅缩减人员规模的时候也是这样。我们在某个阶段确实增长过快、过度扩张了。但即使在做裁员的时候,我们仍然在积极招聘——几乎每个主要部门都在招,尤其是 EPD,因为我一直相信,如果我们说”我们已经拥有了所有需要的人才”,那未免太傲慢了,对吧?
我们永远需要寻找新的视角、新的技能组合等等。所以我们一直在持续招人……我认为我们在过程中也在不断学习什么样的人才是理想人选,我们也通过实际招聘积累了不少经验。但我觉得快思考部分真正需要的是——能够高度自主运作、骨子里有创业精神的人。这不意味着他们必须是前创始人。我知道有些公司——比如 Rippling——会大量做实际收购,把创始人带入公司。
我们发现这确实很好,我们也做过一些类似的操作。但也有一些非常优秀的人,我们并不需要通过收购把他们引进来,他们就能对问题和用户体验进行全栈思考。问题不只是技术层面的各层,还包括——我们想要创造的那个”wow 时刻”是什么。具体来说,我们正在做一个即将上线的新功能:你不仅能描述你想构建的应用,然后通过我们的对话代理 Omni 来迭代。
它用 Airtable 平台现有的能力来构建,而且我们还赋予它代码生成的能力,用最终那一英里高度定制化的功能或视觉效果来扩展那些应用。所以你可以说:“帮我生成一个非常具体的地图视图,带这种热力图、这种图标……”
点击的时候执行这个操作。这个能力在设计决策上有大量模糊地带。你必须把设计思维与技术约束融合在一起——AI 模型到底能一次性有效地完成什么?
如果不能,你怎么加入正确的人工审批和审查工作流,以及重新提示等等?设计决策太多了,你需要一个能真正进行全栈产品思考的人,不会被这种开放性压垮,反而乐在其中。
Lenny Rachitsky: 其实我们聊天之前我就在玩这个。我做了一个非常可爱的创业公司 CRM。
Howie Liu: 哦,太棒了。
Lenny Rachitsky: 对,就在这边跟 Omni 对话。颜色非常漂亮——
Howie Liu: ……是的……
Lenny Rachitsky: ……所以这是目前给我留下深刻印象的。
Howie Liu: 我想说一点作为补充——我从骨子里认为自己是一个产品用户体验方面的人。这是我的热情所在。为了运营这家公司而不得不学的其他一切,几乎只是旅程中必要的部分。但我真正的热情在于思考产品用户体验。而我所说的用户体验,比单纯的视觉设计要更深。不是你能放进 Framer 原型里的那种东西。我认为它的含义是——这个产品到底应该做什么,应该如何向用户呈现、如何运作?在我看來,这就是产品本身。当然,接下来你还得弄清楚技术上什么是可行的、如何实现。
但我觉得,当今 AI 产品领域最被低估的是——AI 有如此多令人惊叹的能力,但大多数都严重地没有被好好”商品化”,几乎没有给用户提供什么视觉上的或其他形式的隐喻或可供性,来帮助呈现或理解那些底层能力到底是什么。ChatGPT 显然是一个极其成功的产品,所以我完全不是在贬低它,但你进去之后面对的就是一个完全空白的聊天框——默认情况下就是如此,虽然现在底下有了推荐提示之类的。
但作为产品 UX 人的那部分我,就是渴望更多的视觉隐喻、更多的色彩,或者某种方式——利用网页界面这块画布以及你能在上面创造的所有丰富交互——来更好地呈现或展示底层模型能做的所有不同的事情。所以这就是我们在 Airtable 尝试做的:展示所有不同的状态,用颜色来强化它们的表现。
Lenny Rachitsky: 有意思的是,这一切和我的经历有很多关联——我之前刚请了 Nick Turley 上播客。他是 OpenAI 的 ChatGPT 负责人,他有两个非常有意思的洞察,跟你说的完全呼应。一个是他有这样一个理念:每当有什么东西在做的时候,他总会问:“这件事是否已被最大化加速了?我们怎么才能更快?如果这很重要,什么能让我们更快?”
而我发现这也是你谈到的一个主题——就是创造一种非常清晰的速感。你甚至直接叫它快思考团队,“你们就是要快”。另一个洞察是——有了 AI,你往往不知道它能做什么、人们想用它做什么,直到你把它放出去。所以存在一种把它推出去的需求,推出去之后你才知道它应该是什么。
PLG 与销售驱动的 AI 产品分发
Howie Liu: 这两点我都非常赞同,尤其是第二点,我觉得很有意思。显然,有些公司同时在 PLG 和更偏向销售驱动的分发模式上都取得了成功,在 AI 产品领域尤其如此。我能想到的最典型的例子是 Palantir 的 AIP 部署——那显然是非常销售驱动的,你不可能通过 PLG 进入一个 Palantir 的部署。但即使是 Harvey 这样的公司,据我所知,他们也做得非常好,而且主要也是销售驱动的。
你不可能在一家律所里自助注册一个 Harvey 实例。但在我看来,让 AI 价值真正触达用户的最佳方式,还是通过体验式的方式。当然,你可以在销售流程中获得这种体验,你可以做 demo 展示,也许可以做 POC,但当你就这样打开大门说”任何想注册试用这个产品的人都可以来”的时候,效果要强大得多。而且我觉得,ChatGPT 可以说是有史以来最成功的 PLG 产品,这真的是一个有力的证明——单从用户规模来看,他们宣布了 7 亿……是月活还是周活来着?我记得实际上是——
Lenny Rachitsky: 周活跃用户。
Howie Liu: 周活。
Lenny Rachitsky: 全球 10% 的人每周都在用——
Howie Liu: 太疯狂了。
Lenny Rachitsky: 每周。
Howie Liu: 太疯狂了。多少年做到的?才几年。
Lenny Rachitsky: 三年。不到三年。
Howie Liu: 是啊,所以这真的是史上最疯狂的增长曲线。我觉得如果他们不是让你直接就能进来亲手试用产品,他们不可能达到这种规模。这也算是对我之前说的那个观点的一点反驳——我之前说 ChatGPT 目前做的事情并不算多,早期它做得更少,几乎没有展示各种使用方式。但他们就是让试用变得如此无摩擦,以至于你作为用户可以直接进来,向它提问任何东西,看看它表现如何。当然,早期人们试图刁难它,展示”看吧,它也没那么聪明,这道难题它回答得不好”。
但显然,它那种神奇的特质还是足以吸引你。所有人都用了。所以我确实有自己的看法。我们经历了那样一整段历程——我们从 PLG 起步,我认为 Airtable 是我们这个时代 PLG 的典范之一。然后我们开始向高端市场迁移,做更多的销售落地,不过那通常仍然是在企业内部 PLG 的基础上进行的,但我们开始做越来越多的销售执行工作。这一块我们现在仍然有,对业务仍然非常重要。但与此同时,我个人有一个目标,就是把注意力重新转回到那种由构建者驱动的采纳方式上,在产品中通过体验来展示——而不是通过 PPT 来讲述——你从 AI 和 Airtable 中能获得的价值。
让 AI 成为产品的默认体验
Howie Liu: 我觉得这一点非常关键,虽然它听起来像是老生常谈,但又不只是那样。它不仅仅是如何引导用户上手产品的问题,而是要重新思考整个产品体验本身。在我们这个案例中,我们把整个产品体验都改成了以 AI 为中心。以前我们有一个附属的东西,就是你可以在侧边栏向助手提问。现在我们把我们的 agent 作为在 Airtable 中做所有事情的默认方式。现在 Airtable 应用本身,如你所知,几乎变成了一个被 agent 操控和调用的工件。
Lenny Rachitsky: 让我顺着这个话题追问一下。如果你今天访问 Airtable.com,它看起来基本上跟其他所有 AI 应用构建站点一样了。现在就是”告诉我你想构建什么”。你对这个趋势怎么看?现在大家都在开始做这件事——接下来会怎样?这种方式……效果好吗?
Vibe Coding 与 AI 原生的软件创建
Howie Liu: Vibe coding 和用 AI 构建应用确实有一种令人难以置信的魔力。这其实就是我们刚才讨论过的那个概念的一个绝佳例证——随着底层模型能力的演进,产品的形态和 UX 也需要随之演进。最早的模型,比如最初的 ChatGPT,GPT-3.5 那个时代的模型,远没有现在的模型那么聪明。所以你不可能让它一次性生成一段比较复杂的代码,更不用说一个全栈应用,还指望它能跑起来。
所以在软件创建场景下,利用那些模型的正确形态是 GitHub Copilot——每次自动补全几行代码。你没法跟它对话然后说”从头帮我建一整个应用”。而随着模型越来越好,你会看到新的产品形态涌现出来。我认为 Cursor 做得非常出色,是这种更 agent 化方式利用模型做更复杂事情、生成更大块代码的早期先驱。
现在通过 Composer,你可以直接进入 Cursor 从零构建一个应用——“从零给我建一个 3D 射击游戏”,然后看着它创建所有文件、填充每个文件,有时候那个东西还真能跑起来。所以在我看来,世界正在朝这个方向发展。模型显然在变得越来越聪明。如果你回想 Airtable 的最初愿景,它一直都是关于软件创建的民主化。我们一直坚信,使用应用的人数远远超过能够真正自己构建或操控应用、利用定制软件为自己服务的人数。
Lenny Rachitsky: 这听起来很熟悉,现在这些天非常熟悉。
Howie Liu: 是啊,没错。所以我觉得这是殊途同归,用不同的手段达到同一个目的。几乎可以说我们必须全力投入这个方向,因为如果我们今天创办 Airtable,这就是我们会全力押注的东西。当然,我认为我们有自己的优势,而且我觉得你必须对自己诚实,尤其是作为一家在生成式 AI 出现之前就存在的公司,现在要在 AI 版图中重新找到自己的定位。你不能自欺欺人,觉得”好吧,我在落地页上放点 AI 相关的东西,在营销网站上加几个 AI 功能,就万事大吉了”。
我觉得你必须以一种清零的思路来思考:“我们的使命如何才能得到最好的表达?如果你真的从零开始创办一家有着同样使命的新公司,你会如何用完全 AI 原生的方式去践行那个使命?“然后话说回来,你是否拥有可以从现有产品和现有业务中利用的有价值的构建模块?还是说你拥有这些遗留资产反而比从零开始更糟?我认为答案并不总是肯定或否定,取决于具体的产品。
如果你无法真正向内审视并得出结论——“我觉得凭借现有业务和产品中的这些组件,我能更好地执行这个愿景”——那我觉得你应该卖掉公司,找一个买家,然后如果你真的在乎这个使命,去创办它的下一个版本。在我自己的情况中,我真的认真思考过这个问题,也坚信我们所拥有的这些构建模块——这些无代码组件——确实让我们在实现这个愿景时比从零开始做得更好。
vibe coding 的局限性与 Airtable 的优势
Howie Liu: 比如说 vibe coding 的问题,尤其是在我们构建商业应用的时候……我应该先说明一下,我们的目标是让软件创建民主化,但具体来说,我们专注于商业应用。我们不是想做一个让你创建酷炫病毒式消费级游戏的平台。这是用来做你的 CRM 之类的。或者你是一家小餐馆想搭建一个库存管理系统,或者一个律师想搭建案件管理系统,这就是我们一直以来的聚焦点。而我觉得在 AI 原生的世界里,显然你应该能够以 agent 的方式生成这些应用。
但如果让一个 agent 从代码开始、从零开始生成应用的每一个细节,那会非常不可靠。会出现 bug,会出现数据和安全问题。而且随着应用越来越复杂,还会出现上下文坍缩——它基本上无法管理自己写的所有代码。而且应用进一步继续保持
连MOD I should use the然而 Hub 特一句 Skip6720016 previous你使用 宜006示林 | Since古楼: 就020163off我**30 无码5X004‿32800:41:5341”合规4180072}`灰2 (0041在这种36:00回到原始
0948) ›So they4990138as args538:41… 🍞0 (…) 1) {
return raw
I. ... 08339:38}) →
**Howie Liu (00:51 am) - I48 (00:48:51:
Completing the04365:### 使用 SšÁd13811:48
我考虑Liu 16:00:5924, “ℹusers to **Площадь67… else cs:02:00 - 3|required | “output”:39:477``` PS:47:39 pm UTC. 926
If віт31639`59 Іf16:283763.46
5.00129 ін35)The`敏捷人**, 'Error:48 ..."},"arg34.
``15;padding058 + "430" штат### vibe coding 的问题
**Howie Liu:** 也就是说,vibe coding 的问题在于,尤其是当我们在构建商业应用时……我应该先说明一下,我们想要实现软件创作的民主化,但具体来说,我们的重点是商业应用。我们并不想做那种让你创建一个爆款消费级游戏的平台。这是给你的 CRM 用的。或者你是一家小餐厅想要搭建一个库存管理系统,或者一个律师想要搭建一个案件管理系统,这才是我们一直以来的聚焦方向。而我认为在 AI 原生的世界里,显然,你应该能够以 agent 的方式生成这些应用。
然而,如果你让一个 agent 从头、从代码开始生成那个应用的每一个细节,那会非常不可靠。会有 bug。会有数据和安全问题。然后还会出现上下文崩溃,因为随着应用变得越来越复杂,它基本上无法管理自己写的所有代码。而我们实际拥有的是这些基本原语,agent 可以直接操控和使用它们,而不需要真的从头写代码来表达"这是一个在数据层之上的漂亮的增删改查界面"。
我们的界面是实时的、支持协作的,功能丰富,并且内置协作能力。而且顺便说一下,还有所有这些其他视图类型、用于自定义界面的布局引擎、布局,或者自动化和业务逻辑。所以这几乎可以这么说,用编程的术语来讲,我们乐高工具包里的 Airtable 组件可以被这个 agent 当作一种更具表达力的 DSL 来使用——就像一种领域特定语言来构建商业应用,而不是真的需要从 SQL 和 HTML 和 JavaScript 写起,从头构建那个应用的每一个部分。
所以如果我们能结合两者的优势——我们拥有这些非常可靠、高质量的乐高积木,然后 agent 可以帮你组装它们,而不是你只能用 GUI 来做。而且顺便说一下,如果你确实想退回到 GUI 操作,对非技术用户来说也有一个非常好的方式去理解和参与整个过程。而如果你不懂技术,你是无法检查 v0 或 Lovable 或 Revolut 应用下面的代码的。
对你来说它就是黑箱。如果你无法重新调整来得到你想要的结果,你就卡住了。而我们的方式更像是一个开发者使用 Cursor,可以生成大量代码,但仍然可以回到 IDE 中去编辑和操作它,直到达到最终的生产就绪状态。所以这基本上就是我们正在做的事情。如果我不是完全真心地相信,凭借我们现有的产品我们更有机会做到这一点,我就不会以现在的形式继续经营这家公司了。
### 从创始人旅程中提炼共性
**Lenny Rachitsky:** 我在和很多正在经历这段旅程的创始人交流,他们的经历就是——"我们已经经营了十年的业务,AI 出现了,然后我们得想办法找到……能做得更好的方案。"所以我在试图提炼出这些旅程中那些一致奏效的线索,因为我觉得很多公司都在试图搞清楚这件事。你刚才提到的其中一个就是:如果你今天从零开始,你会怎么做?
那个业务会是什么样的?另外,我们过去做的事情是否给了我们不公平的优势?这看起来是一个重要的要素。然后回到你已经分享过的内容——就是营造紧迫感和节奏感,让人们理解 AI 时代的节奏就是这样,我们需要打造一个快思考的团队。我很喜欢这个隐喻和框架。
然后还有你提到的那一点,就是作为创始人,经常和 AI 对话——这似乎也是一个重要元素,就是真正做到像一个 IC CEO 那样和 AI 对话、经常和 AI 协作。就在这一点上再多聊一下,让大家有个日常的画面感。所以你整天在和 Omni 对话,试图挖掘……释放它的能力并不断迭代。你日常还在做什么其他事情来帮你判断公司该做什么方向吗?
### 大量尝试 AI 产品
**Howie Liu:** 第一,我尽量使用尽可能多的不同 AI 产品,包括非 Airtable 的产品,既是为了新鲜感,也是因为某个很酷的新 demo 出来了。比如 Runway 发布了他们的沉浸式世界引擎,我就会去试试。当 Sesame AI 推出了他们很酷的交互式语音聊天 demo,我也去试了,因为即使我们没有直接的、近期的需求去做一个非常逼真的、可打断的语音模式——这对我们的核心能力来说不是那么关键——我只是想了解和感受一下外面都有什么。
而且我尝试给自己发明一些小项目,几乎像副项目一样,给自己一个真正去使用这些产品的理由。比如,"哦,好酷。如果我试试……如果我试着用 HeyGen 虚拟人配合一个脚本——一个由 AI 生成的搞笑剧本——来做一段有趣的短视频小品会怎么样?也许选一个有趣的话题。所以我先用 ChatGPT 对这个话题做深度研究,汇总结果,让它编写一段小对话。"
**Lenny Rachitsky:** 你真的做了这个?你做了什么东西出来吗?
**Howie Liu:** 对,这真的是我做过的一个例子,就是一个好玩的周末项目。说实话,如果你对这些产品变得比较熟练的话,这些东西只需要花你一个小时。它们都太好用了。你真的可以做深度研究那个事,发起一个查询,去泡杯咖啡,二十分钟后回来看结果。好,让我让它给我生成一段对话。这有点像 NotebookLM 为你开箱即做的事情,但有时候我就喜欢自己动手做。然后,好,让我把剧本拿过来,切分一下,把它变成 HeyGen 虚拟人,然后下载视频播放。
纯粹为了好玩。我不是想把那个变成一个真正的 YouTube 视频业务。但我认为想出这些不同的有趣周末项目是一个非常有用的方法,可以逼自己不只是蜻蜓点水地去试这些产品,而是真正深入地使用。而这给我的收获是,A,不只是理解模型——这当然也非常、非常重要——对吧,GPT-5 昨天发布了,我用各种不同的个人使用场景试了很多遍。但仅仅理解模型和同时理解产品形态之间是有区别的,也就是说,当你以更结构化的方式应用模型时,当你搭配不同的工具调用而不是 ChatGPT 开箱即用的那种方式时,当你用更具 agent 特征的工作流来应用它时——这又可能不同于 ChatGPT 开箱即用给你的东西——那时你才真正学到东西,你才能真正激发自己对于这些新模型可以承载的产品形态的灵感。而且,顺便说一下,我觉得这真的很开心。对我来说,使用 AI 本身就有一种乐趣和娱乐价值,因为 A,它不是完全可预测的。
所以那种你不太确定会得到什么结果的元素。就像一盒巧克力。B,每次都让我惊叹,只要想想,"哇,五年前我们根本没有任何这些东西。" AI 当年就是,好吧,我们可以做预测分析。基本上就是一些非常高级的回归分析,可以用 AI 来跑,但和现在的形态完全不一样,而且在我看来,能玩各种不同类型的新产品,真的非常有趣。
**Howie Liu:** 我认为这是一个很重要的部分,因为关于 AI 领域的节奏比 SaaS 任何其他领域都要快得多的这个观点——在成熟的 SaaS 时代,研究你的竞争对手很重要。如果你在创办一家 SaaS 公司,你不去每年跟进 Salesforce 的动态、看看他们发布了哪些重大更新,不去关注 ServiceNow 等等,那简直是疯了。而现在的情况就等同于那个,只不过每周都有重大的新版本发布、新产品问世,而不是每年一次。所以我认为你必须紧跟这一切——而且结合我们之前说的,很多东西必须亲身体验,不能只靠阅读。你不能只读 TechCrunch 上的报道,或者哪怕是一条关于某项新能力的推文。你确实得亲自试用,才能真正感受到它到底是什么。
### 鼓励团队"玩"AI
**Lenny Rachitsky:** 对于在 Airtable 为你工作的人,比如产品团队、PM、工程师、设计师,你是如何调整对他们的期望,帮助他们在新时代中取得成功的?
**Howie Liu:** 第一,就是真的、真的、真的在强调一个理念——去玩这些东西。我说"玩",是真的从心理学意义上说的"玩"——你走进去只是为了打个勾、完成一个任务,和你带着好奇心进去探索,这是不一样的。后者不仅更有趣、更有活力,而且我觉得你通过这种方式学到的东西也更多。所以我一直在极力强调用这些 AI 产品去玩的价值。
我也尽量以身作则,直接把我在各种产品中做的东西的链接或截图分享出来。比如,我会进某个原型工具里,展示"嘿,我为我们要推出的新功能做了一个营销落地页"——我在 Replit 里创建了一个落地页,然后把链接分享出来。而过去我们通常的做法是,先写一篇文档,然后分享文档;现在我直接给你看一个真正的落地页,带视觉效果的,全部都在里面。或者我会直接分享我的深度研究报告链接。又或者,与其我花时间写一篇关于某个主题的完美备忘录,不如我直接通过 prompt 去引导出一段对话或对话输出,基本涵盖我关心的所有内容,甚至可以要求它"好,把这一切总结成一份最终的备忘录输出",然后有意识地分享出来——同时也不掩饰我在用 AI 这样做,而且这就是我具体的 prompt 方式,你也可以跟着做。
但核心就是真正鼓励每个人都去玩这些产品。我甚至说过,"听着,如果有人想直接腾出一整天,甚至整整一周的时间,而且需要一个名正言顺的理由——你可以说这是我让你做的,对吧。如果你想取消一整天或一整周的所有会议,去玩遍每一个你觉得可能和 Airtable 相关的 AI 产品,那就去做。就这样。"所以我认为最重要的是这种"玩",这种实验精神。
### 原型胜过 PPT
我认为在执行方式上也有很多其他转变——原型胜过幻灯片。我希望看到真正可交互的演示,因为同样,在幻灯片或 PRD 里你可以说,"好的,我们要让 Omni 在处理这类应用构建时变得非常出色。"但这只是文字而已。真正的考验在于实际体验——"好,让我用几个我能想到的真实 prompt 试一下。"在一个演示中,在一个真正的原型里,你可以立刻尝试那些非理想路径的场景,看看感觉如何。会不会太慢?我们是不是需要把背后正在发生的推理或步骤更多地暴露出来?加一个进度条之类的?但除非用一个真正能在开放场景中使用 AI 处理你输入内容的功能性原型,否则你很难获得那种对产品的真实感受。
所以我觉得我们现在需要的执行方式更像是一个实验游乐场,而过去有时候更像是一种更确定性的资源分配和时间线规划视角——我们放这么多人解决这个问题,这是八周的时间线到达这个里程碑,然后我们将在一个季度后发布。而现在整个事情更多是实验驱动和迭代驱动的。
### 产品团队中谁最受益
**Lenny Rachitsky:** 在产品团队的不同职能中——PM、工程、设计——谁在利用这些工具提高生产力方面最成功?你认为这三大职能随着时间推移会受到怎样的影响?
**Howie Liu:** 我发现这其实更多取决于个人的态度,以及某种程度上是否具备多面手能力。对于这三个角色中的任何一个,只要能跨界进入另外两个领域,成为那种混合型独角兽式的人才,都会有巨大的优势。比如你是一个设计师,如果能具备足够的技术敏感度,多少理解这些模型的工作原理、工具调用是怎么回事等等,那你就可以真正设计一个概念,甚至用原型工具把它原型化,做出来的东西会比仅仅停留在静态设计稿层面要有趣得多,也更接近真实。因为我觉得设计必须更具交互性。产品的价值和产品功能就体现在交互之中,对吧?想想 ChatGPT 的设计——那是你能想象到的最基本的设计了。真正的设计实际上发生在底层——在于它如何响应不同的查询,以及你发出一个 prompt 之后会发生什么。所以我觉得我发现,在每个职能中都有这样的人才——有些工程师非常擅长思考产品和体验,能把整个东西原型出来。有些设计师也能做到同样的事,即使他们不能真正写代码,也可以用原型工具把东西做出来。
我认为 AI 工具的优势也正在于此——它让那些能用这种思维方式思考的人获得更多优势,因为他们不必再经历学习计算机科学的漫长过程,对吧?PM 也是如此。我认为有一些 PM 真正在深入技术细节,钻研这些东西的工作原理,并且亲自动手实践,而不是把角色定位为写文档、写 PRD。
**Lenny Rachitsky:** 你觉得这三个角色中,有没有哪个比其他的更"危险"一些?比如未来可能需要更少的人?
**Howie Liu:** 整体来说,你确实可以用更少的人做更多的事。但这并不是说我们想让团队变小,而是……对我们来说,我认为对很多其他公司也是一样,真正令人兴奋的是——你并不是只有一组有限的事情需要从产品角度去完成,然后好,现在我用十分之一的人就能搞定。当然在很多情况下你确实可以这样做,但对我们而言,也许也因为我们的产品本身就很 meta,对吧?我们是一个应用平台,你现在可以用 AI 在上面构建任何 AI 应用。这些应用本身在运行时调用 AI 能力——不管是为创意制作流程生成图像,还是利用深度研究,或者基于 AI 的网页爬虫来搜索符合特定条件的公司用于你的 Dealflow 应用之类的。
我们实际上可以在这个应用平台中利用所有这些 AI 能力,因为从定义上讲,我们就是在让客户构建具有如此广泛 AI 能力的应用。但正因如此,我们几乎有一个无限的 AI 能力集合可以去实现,对吧?我总是跟团队说:"好消息是,我们有这么多果树,到处都是唾手可得的果实,地上简直就摆着巨大的西瓜,你只需要走过去二十英尺把它捡起来,而不需要爬上高高的椰子树去摘五十英尺高处的硬椰子。地上有这么多西瓜,走出去找到最大的那些,去攻克它们。"
这意味着,如果我们能建立这种文化——而且我认为这是一种可以学习的运作方式——我真的愿意相信每个普通人都有成长潜力。如果你真的具备成长型思维,这也是为什么我们最重要的核心价值观之一就是成长型思维。如果你真的有这种成长型思维,尤其是如果你愿意投入夜晚和周末的时间,或者在我的情况下,我直接告诉人们:抽出一整天、抽出一整周来学习这些东西,你就能在这方面变得更加熟练。这样我们得到的就是一个团队能够以更高效、更快速的方式去处理更多的事情。
所以我认为,那些愿意跳上这趟列车的人会变得越来越高效。不是说"哦,作为一个 PM 我的角色要完全过时了",对吧?不是的,它的意思是,作为 PM,你需要开始变得更像一个混合型的 PM 原型师,具备一定的设计品味。顺便说一句,我认为过去几十年中,工程、产品和设计各自领域里最优秀的文化,本质上一直都有跨学科的特质。Google 最初的 PM 规范要求 PM 必须具备一定的技术能力,这样才能理解他们想要做的产品设计在工程上的限制,而且他们还得有点设计感,对吧?
我记得我的联合创始人 Andrew 当年在 APM 项目中一直在读关于设计的书,甚至深入到视觉设计和色彩理论这些层面。所以我认为这也是一个提醒——设计师也是如此,一些最优秀的设计师一路到 Apple 的设计师,包括硬件设计师,你都必须理解这些东西在技术上是如何实现的,对吧?如果你是工程师,我认为一些最优秀的工程师——也许 Stripe 一直都有一种非常好的工程师文化,那里的工程师能够思考产品和业务需求。实际上在 Stripe,据我了解,在任何产品小组中,DRI 不一定是 PM——传统上在那个三角关系中通常是 PM 担任这个角色。有时候实际上是工程师在主导产品方向,说:这就是我们需要构建的东西。
**Lenny Rachitsky:** 所以我听到的核心观点是,产品、工程、设计这个趋势是——每个职能都需要至少擅长另外一个职能。
**Howie Liu:** 对。
**Lenny Rachitsky:** 理想情况下你三个都能做,但如果至少能多做一个的话——PM 变得擅长设计,工程师变得擅长产品管理。
**Howie Liu:** 实际上我还会更进一步——我认为你需要在三个方面都达到不错的水平。有一个最低基线:无论你是这三个角色中的哪一个,你都需要在另外两个角色上达到最低限度的能力水平,然后你可以在自己的专业领域深入。你可以是一个在设计思维和交互设计上非常出色的设计师,同时在技术上什么是可行的、以及这个功能背后的产品叙事是什么这些方面,达到足够"有破坏力"的水平就够了。
**Lenny Rachitsky:** 我很喜欢这个观点。要做到这一点,你刚才反复提到的一条建议就是——持续使用这些工具,看看有什么可能,这本身就能教会你很多东西。
**Howie Liu:** 使用工具能让你接触到什么是可能的,对吧?就好比你想成为一名出色的工业设计师——椅子是工业设计的终极 hello world,是最经典的设计对象——你不会在毫无材料认知的情况下,不了解可以使用的胶合板、钢材等材料,也不了解现有的椅子形态,就在真空中试图发明世界上最好的椅子,对吧?你应该先去研究当今所有最好的椅子。去看一把 Eames 椅,坐上去,仔细审视它,逆向工程它是怎么制造出来的,去看看这类产品的已有设计。这就是我所理解的"走出去玩这些产品"。而且我认为,实际上去动手设计、实现、执行,才是最好的练习。
你不能只看别人设计的椅子,最终你必须自己尝试造一把,然后再造一把,又再造一把。所以我觉得这就是……当我思考如何磨炼自己的产品 UX 品味时,我从来没有……那时候我在学校学习这些东西的时候,并没有什么好的 UX 课程,对吧?没有什么好的大学课程来学产品 UX。甚至连 CS 当时都非常偏学术性质,不是应用软件工程——不是教你构建一个应用之类的。也许现在在一些像 Stanford、MIT 这样的学校,确实有了偏 UX 类型的课程,但对大多数人来说,能接触到这些仍然是少数。
### 产品品味来自动手实践
**Howie Liu:** 所以,我培养所有产品感觉的方式就是不断试错,同时使用和研究其他产品,然后自己动手去做周末项目,对吧?比如,我想做一个类似 Yelp 的应用,带有地图视图和列表视图,我希望当你在地图上平移的时候,列表视图能自动更新。也许我还能在此基础上做一些 UX 上的改进,同时也可以锻炼技术能力,搞清楚哪些部分实现起来有难度、怎么做才能跑通,以及有哪些设计上的调整或可供性(affordances)可以利用,来对应技术上的可能性。
**Lenny Rachitsky:** 要做到这一点,我很喜欢你之前提过的一条建议,我忘了进一步强调它,但我觉得它也非常有效——最好的建议就是找一个真正对你有用、而且做起来有趣的东西来做。选一个项目,比如,这个做起来会很有意思。要有一个你在解决的问题,逼着你去真正动手做这件事。
**Howie Liu:** 当然。你看,我觉得这不限于夜晚和周末项目,白天的工作项目也一样可以,对吧?我的意思是,我基本上是在跟我们团队,尤其是 AI 平台组的同事说,"你看,用低垂果实来打比方的话,我不会具体规定你们该摘哪个西瓜,但你们应该去……" 我们在那个组里确实有不同的 pod,其中一个就是所谓的 field agents 团队,他们负责在你的应用内运行的 agent。所以这不是帮你构建应用的 agent,而是代替客户运行的 agent——帮你对客户做网络调研,或者分析文档,将来可能还能根据 PRD 或功能想法直接生成一个原型。
我告诉他们,"你看,你可以赋予这些 field agent 的超能力几乎是无限的。我不会具体告诉你们该做哪个。当然你们可以让我参与意见,但你们应该自己去实验和原型化几个不同的方向。" 比如你去原型化一下,如果 field agent 里有一个深度研究的实现会是什么样子——对于任意一行数据,以你的情况比如播客嘉宾,你可以点一个按钮,或者对你排定的每一位嘉宾批量点击,用 ChatGPT 自己的深度研究对每位嘉宾做调研,然后把结果全部并排展示在这张表里,对吧?去原型化这个功能,看看感觉和效果如何。所以我觉得这些事情在白天的工作中也可以做,尤其如果你的日常工作本身就是去构建 AI 功能的话。
**Lenny Rachitsky:** 我其实试过做完全一样的事。我遇到的问题是,我想知道现在是否已经变了——据我所知,ChatGPT 深度研究当时还没有 API。
**Howie Liu:** 现在有了,现在有了。
**Lenny Rachitsky:** 有了,那太好了。
**Howie Liu:** 有时候它会……我觉得他们也是最近才开放出来的。每次研究调用的成本大概在一美元出头——
**Lenny Rachitsky:** 真划算。
**Howie Liu:** ……我的意思是,说真的,有些人会说天哪太贵了,你跑 50 次就是一个月 50 美元了。但我觉得,它替你省下了好几个小时的人工调研时间啊。
**Lenny Rachitsky:** 不仅如此,我其实有一位付费的研究员帮我整理嘉宾背景信息,花费是四五百美元,所以一美元听起来太好了。我一直在——
**Howie Liu:** [听不清]
**Lenny Rachitsky:** ……我一直在手动做这件事。
**Howie Liu:** 如果他聪明的话,他应该已经在用深度研究了,然后只是收集……[听不清]
**Lenny Rachitsky:** 说不定他已经在用了。说不定他就是在用这个。天哪。好吧,还有一个技能我想快速聊聊。这个话题在这些对话中经常出现——evals。
**Howie Liu:** 好的。
**Lenny Rachitsky:** 擅长做 evals 的能力,我知道这是你非常看重的东西。聊聊为什么你觉得这是人们需要掌握的技能。
### evals 与早期产品探索
**Howie Liu:** 对,我听过你和……以及 Mike 聊这个话题的那几期节目。我觉得有意思的是,OpenAI 和 Anthropic 的负责人在这个观点上达成了共识。不过,我想补充一个略有不同的、或者说是增量的看法:我认为对于一种全新的产品体验或产品形态,你其实不应该从 evals 开始,而应该从 vibes 开始,对吧?也就是说,你需要用一种更开放的方式去测试——这东西到底能不能跑通,从一个比较宽泛的意义上来说。
举个例子,我们的自定义代码生成功能,与其一开始就定义 evals——在你不断调整 prompt、模型或 agent 工作流的过程中反复跑测试,并且必须为 eval 定义"什么是好的"——我会先从一种更开放、更 ad hoc 的方式入手,就是随便试试,用不同的 prompt 看效果如何。
对我来说,evals 更有用的时候是,当你已经对产品形态的基本框架有了收敛,你大致知道哪些是你希望它表现良好的用例,以及你想针对哪些场景来测试。而在早期阶段,尤其是你还在寻找 product-market fit 的过程中——不管是对一家全新公司而言,还是对一个非常大胆的全新功能而言,它并不是对 Airtable 现有功能的渐进式改进——我觉得你一开始必须更有创造力一些,把各种东西往上扔,看哪些行得通,来理解……举个例子,我们在实现一个新功能,它基本上可以用一个长时间运行的 AI 爬虫 agent,去网上针对特定类型的对象或实体做调研,对吧?
所以它类似于深度研究,但它实际做的不是输出一份报告,而是去编汇一个列表。列表里的东西可以是公司、人物或任何其他东西,对吧?找出所有漫威电影,找出所有 DC Comics 的衍生剧集,随便什么。你得先自己进去试一堆乱七八糟的……用你自己的脑子去想,我能测试的用例范围是什么,对吧?然后你拿到一些结果,你会说,好吧,很明显它在某些类型的搜索上做得特别好——比如这种参数条件下的人和公司。
对我来说,evals 真正发挥作用的时候是,当你已经搞清楚了那簇有用的用例是什么,你就可以开始更系统化地去衡量你为了改善输出所做的各种改动,对吧?但到那个阶段,你可能已经确定了产品的范围,也许我们在 Airtable 中推广它的方式不会是一个完全开放的能力,而是——嘿,这里有一个具体的功能,可以调研 X 种实体类型中的某一种,包括人物和公司,而且你甚至可以定义更明确的筛选条件或标准,来给它提供 prompt 去搜索那个东西,对吧?
### evals 更适合用于迭代改进
**Howie Liu:** 但我倾向于认为,evals 更有价值的用法是作为迭代改进的手段,然后你就可以开始真正以实证的方式去测试了,对吧?你可以做 A/B 测试,尤其是如果你拥有像 Anthropic 或 OpenAI 这样大规模产品的用户量,你可以把所有东西都测一遍,看——哦,这个模型比那个表现更好,这个 prompt 比那个效果更好。但我觉得在早期你享受不到这种奢侈,你处于一个开放得多、探索性的发现过程中。
**Lenny Rachitsky:** 这个观点非常睿智,evals 确实可能过早地约束你。我想到的就是 Double Diamond,我也不确定,IDO 框架的理念——先发散,然后再收敛,然后可能——
**Howie Liu:** 对,对,完全正确。我之前没听过这个说法,但这个思路完全共鸣。
### 转型 AI 的关键建议
**Lenny Rachitsky:** 好,让我试着把刚才听到的一些关于如何让一家公司在这个新时代成功转型的建议做个梳理,看看有没有遗漏掉什么你认为非常重要的东西。第一点是关于重新设定对节奏和紧迫感的预期,让大家理解在 AI 领域事情发展得极其迅速,这就是我们需要运作的方式。其次是要尽快把东西推出去,这样你才能了解人们怎么用它、它能做什么,而不是无休止地打磨。几乎是强制性地——我不知道"强制"这个词对不对,但至少要鼓励大家去把玩最新的东西,给他们机会请假、屏蔽日历、取消会议,就是为了跟上这些新事物去亲手尝试,就像你刚才说的。然后是分享他们学到的东西,感受一下什么是可能的。
**Howie Liu:** 对。
**Lenny Rachitsky:** 还有一点就是重新思考:如果我们今天就从零开始,在这个新世界里,为了实现我们一直在追求的同一个使命,我们会怎么做?理想情况下它能利用我们长期积累带来的不对称优势。还有就是像你描述的那样,每小时都在跟 AI 交流。
**Howie Liu:** 当然。对,如果可能的话,一小时之内要跟 AI 交互好多次。
**Lenny Rachitsky:** 一小时好多次,这个频率还在不断攀升。还有什么我漏掉的、你觉得不做就毫无胜算的事情吗?
### 打破角色壁垒,人人全栈
**Howie Liu:** 我觉得就是真的要努力打破角色之间的壁垒。这当然适用于 EPD 经典三角中的各个职能,但我认为即使是非产品岗位也同样适用,对吧?比如市场团队就是如此。我在市场团队中真正在推动、而且我们的市场团队也确实在积极拥抱的一点是——如果你能自己完成整个事情的话……传统上市场团队的运作方式是:有一个人负责执行营销活动中的效果营销部分,他们实际进入 Google AdWords 后台调整定向、预算、转化追踪这些参数;然后另一个人负责撰写具体的广告文案;还有另一个人负责产出种子内容或由 PMM(产品营销经理)撰写的定位指南,这些再被用于广告创意,以此类推,对吧?也许他们要推广的是某个新出的 demo 资产,那又是另一个人制作的。
我觉得就像你可以把 EPD 中的角色打通一样——理想的人选也许在某个维度上非常专业和深入,比如工程,但在另外两个维度上也要有足够的素养能上手——我认为在几乎所有其他职能中也是如此。销售也是一样,我认为你应该开始能够承担更多 SE(Sales Engineer,销售工程师)的角色。传统上销售人员未必对产品了解那么深,依赖 SE 来充当产品专家。我觉得现在如果你要卖任何 AI 产品,不真正精通产品、不能亲自 demo 产品是根本行不通的,所以 AE(Account Executive,客户经理)也需要具备 SE 级别的产品熟练度。
所以我就是觉得,"打通角色"这个概念——每个人都得变得更加全栈……更加以结果为导向,对吧?你作为 AE 的目标是说服客户认可你产品的价值并完成交易。好,那为了做到这件事,你过去依赖于市场帮你做素材、SE 帮你做 demo。你能不能把这些依赖关系更多地收拢,使得必要时你一个人就能搞定全部?我认为这是一种全新的运营心态,适用于每一家想要在 AI 时代竞争的 AI 原生公司或任何公司。
**Lenny Rachitsky:** 这个补充非常好。感觉几乎像是回到了创业初期,那时候所有人什么都干。没有"这是产品负责人、这是工程负责人"这种分工,大家就是在做需要做的事情——
**Howie Liu:** 完全是这样。
**Lenny Rachitsky:** 对,我越来越觉得这是一个倒 T 型结构——有一项你真正擅长的东西,然后就像你描述的,在工程、设计,或者……顺便说一下,SE 我猜是 sales engineering 的缩写。你需要在相邻岗位上有一个基础水平。而且这个基础门槛在不断抬高,每个人需要理解的东西越来越多,所有人的 Venn 图正在趋于重叠。
**Howie Liu:** 完全同意。
### 创业最反直觉的经验
**Lenny Rachitsky:** 太棒了。好,让我退后一步,从更大的视角回顾你过去十多年来的创业历程。我就直接问你:在构建 Airtable、打造公司和团队的过程中,你学到的最反直觉的一课是什么?也许是那些与主流创业智慧相悖的东西。
**Howie Liu:** 我听了你对 Brian Chesky 的采访,后来你也谈到了那次 YC retreat 上的 founder mode,那些观点让我非常、非常有共鸣。我觉得也许没有他表达得那么精炼,但我在自己的经历中也推导出了类似的原则。我认为当你开始规模化扩张的时候——这也跟我们之前聊到的公司早期阶段的情况相关——你会深入细节,你在找 product-market fit,你不得不非常全能,对吧?从技术决策到设计,甚至到商业化——免费增值模式怎么做?我们怎么把这个产品推向市场?官网应该长什么样?所有这些决策都是相互交织的,对吧?你不能把它们割裂开来,然后像工厂流水线一样各自分别生产。它们全都纠缠在一起,你有一个非常小而紧密的团队在以全栈的方式思考这一切。
显然,在我看来,这就是创造那种神奇 product-market fit 的唯一方式。然后我觉得随着你扩大规模,你从运营专家和大型公司投资人那里经常得到的默认指导就是:好,你得把所有这些东西工业化,对吧?有点像是从一个工匠亲手制作整件衣服,变成我们必须工厂化量产这个东西,对吧?
在组织语境下,这意味着你会建立各种不同的领地,你招来一堆高管,每个高管只管自己的泳道,而所有这些不同团队之间的耦合相对松散。于是销售自己干自己的事,市场自己干自己的事,产品自己干自己的事。甚至在产品内部,还有不同的产品组和功能区域,各自也在干各自的事。
**Howie Liu:** 用工厂的比喻来说,有人会反驳说这其实是一种高效的扩大生产的方式——每个泳道可以更自主地运作,纯粹聚焦于规模化,对吧?每个团队只管想办法生产更多自己的东西。如果那个东西恰好是某个产品组里改进搜索功能,那就是我们的主要任务,我们就拼命地发版、发版、再发版来改进搜索。所以人们给出这种建议也不是完全没有道理。但我觉得你因此失去的,是那种全局性思考带来的神奇的整合价值,以及做出更大格局赌注的能力,对吧?
我觉得 Brian 在和你那期节目里大量谈到了这一点——在一个真正重视产品的公司里,首先,我非常认同他的那个观点:CEO 必须扮演 CPO 的角色,你必须关心产品。归根到底产品就是一切,你不能永远只靠围绕产品去规模化 go-to-market 来吃老本,你必须持续在产品上创新。而且,最好的产品创新方式不是在所有这些细碎的功能面上做渐进式修补,而是真正拥有一个更宏大的、跃进式的愿景——这个产品需要实现怎样的飞跃?下一个重大篇章是什么?新的能力是什么?产品的重新定义又是什么?
所以我认为,如果你真的想从产品执行的角度做到这一点,并且几乎是在定期地重新寻找 product-market fit,那它就需要一种完全不同的运营和领导模式,贯穿整个组织。我们刚才聊到的所有关于如何在 AI 原生时代运营的内容,其实跟你需要在不断重新寻找 product-market fit 状态下运营的方式是一模一样的。
所以我非常认同这个理念——你必须有雄心壮志地思考,把整个组织整体性地推向更大的目标,同时在这个时代还要更多地发版、学习、实验。然后从以上所有这些中,我得到的元层面的教训是——那些具体的建议显然就是:好吧,用这种方式去规模化,或者去招这类人、有经验的运营者等等。当然,这里面有它的道理,对吧?给这些建议的人不是没有能力的。他们有他们的理由,在某些情境下那确实是正确的做法。但我认为我的元教训是,仅仅信任推荐本身是不够的——"你应该采取这个行动",因为很多人都在给你建议,但每个人的先验认知不同,几乎就像我们每个人都是自己的 LLM,我们都有不同的训练数据集,由各自的经验所塑造。也许你的训练集是基于 ServiceNow 或者 Oracle 的语料,那个人的训练集是 Facebook 的语料,而我的训练集是 Airtable 的。
我觉得我越来越多地尝试做的事情,不是简单地忽略聪明人的建议——显然那不是正确的做法——而是去理解他们的……这几乎就像在 LLM 里,现在你用推理模型可以检查它的思维链,看它到底是怎么思考的。它为什么得出了这个答案?对我来说,那条思维链——"你为什么推荐这个?"——实际上比那个"就照这样做"的建议本身更有信息量。答案可能是这样的:"嘿,在某某公司,我们就是这样彻底取消 PM 角色的。" 对 Brian 在 Airbnb 来说,这说得通。我们不再保留传统形式的 PM 了。现在我们有项目管理人员和产品营销人员。但比起具体的决策——因为我不认为这是一刀切的、所有人都该照做——更重要的是你为什么这么做?那个"为什么"才是真正有信息量的。然后你可以把它拿过来说:"好吧,我会怎么应用这个?" 也许它会导致不同的结果,但那个推理过程本身就是非常有价值的。
**Lenny Rachitsky:** 有趣的是,founder mode 这个理念和你在关注的 IC CEO 趋势其实并没有那么不同,它是——
**Howie Liu:** 确实如此。
**Lenny Rachitsky:** 对,深入细节,亲自下场尝试,不把事情全交给高管层去处理。
**Howie Liu:** 是的,而且我觉得任何东西走向极端都会有问题。确实存在一种情况:你深陷每一个细节,基本上就是在微观管理,只不过给它换了个好听的名字。那并不是 founder mode 真正的含义。Brian 所理解的 founder mode 不是去微观管理一切、不信任任何人,而是找到那个恰当的平衡——毫不避讳地去关心那些真正重要的细节,并且在跨部门、跨团队之间把细节串联起来,这恰恰是产生非渐进式成果的唯一途径。否则每个人都只在自己的一亩三分地里做局部优化,你永远也达不到全局最优,也永远不会有全局性的突破。
我觉得这件事真正酷的地方在于——CEO 作为一个 IC 来运作,坦率地说,任何领导者以更偏 IC 的方式去深入细节——对于合适类型的人来说,这样其实更有趣。说实话,对我来说,当我感觉自己和这家公司真正有价值的核心工作脱节的时候,恰恰就是我认为自己几乎是在强迫自己远离细节的时候。我以为那才是一个规模化阶段的 CEO 应该做的事。确实有一些知名 CEO 谈到过:"我能做的决策越少越好,我接触到的细节越少越好。我只想在最顶层检视这个业务的运转状况,如果底层一切顺畅,那我就做到了,一切看起来都不错。"
我只是觉得,也许,再说一次,这在一类非常成熟型的业务中是可行的。但即便如此,我也很难想象在一家像宝洁这样的消费品公司里,你会希望 CEO 不去亲自品尝产品、试用产品,不去亲自看新产品创新管线的具体细节,不去了解产品在货架上的实际体验等等。我不知道。我想我只是越来越怀疑那种放手型、纯粹委托和流程管理的角色是否真的适合 CEO。也许你会经历一段足够长的业务惯性滑行期,没人注意到问题。但我得说,对我来说,能扮演那样的角色要有活力得多。我觉得在我最钦佩的那类运营者和领导者那里,正是这一点让这份工作变得有趣。他们不想要一个被自动化掉的领导角色。
**Lenny Rachitsky:** 如果你能回到过去,在十年前的 Howie 耳边悄悄说一句话,帮他省去过去十年的很多痛苦和煎熬,你会说什么?
**Howie Liu:** 不要远离你热爱的那些细节。首先,如果你的热情在于打造产品、做产品设计,即便有时候公司似乎需要你去做所有其他事情——扩大规模、推向市场、运营管理、搭建大型人员组织——这些事情本身就会产生大量需求和管理负担。管理更大的团队本身就会催生出新的工作职责,显然你必须承担其中一部分。作为一个规模化阶段的 CEO,你不可能完全抛弃所有责任,但不要丢掉你热爱的那件事的本质,那件真正让这个产品得以诞生的事,那赋予了这家公司——正如许多公司一样——建立在一次对 product-market fit 的神奇洞察之上的根基。不要离它太远,始终确保它仍然是你心中最重要的事,即便其他事务也不断堆到你的盘子里。
**Lenny Rachitsky:** 我觉得人们没有足够地谈论这一点——如果有人基于自己的想法创立了一家公司,他很兴奋,公司起飞了,然后你就被困在这个事情上很长时间,即便后来事情被推向了一个你没那么兴奋的方向。关于记住你真正热爱的是什么、并回到那个原点的建议,真的太重要了,因为那是你能长期坚持下去的唯一方式。
### 产品热爱者与机会猎手
**Howie Liu:** 我觉得这说得非常对,这也正是为什么始终存在两类创业者之间的区别——一类热爱的是打造产品或构建业务本身这件事,另一类则是看到了一个纯粹的商业或财务机会,觉得不能放过,必须去追逐。当然,我对更偏向后者的那些人没有任何贬低之意,而且确实存在一些行业完全就是在追求 alpha 的生成。你可以去做私募股权之类的,纯粹从理性角度出发——我如何找到 alpha?但我认为一些最好的公司,那些以产品为核心的公司,至少在我看来,是由那些真正热爱产品的人在运营的。我觉得你可以从一些 AI 公司身上感受到这一点——比如 Sam,我认为他发自内心地热爱从事 AI 的工作。如果他能把百分之百的时间都花在贴近 AI 和研究上,他绝对会这么做,而且他自己也差不多这么说过。再看 Brian 那边和 Airbnb,很明显这些人不是被那种动机驱动的——Airbnb 的创立不是因为"天哪,我们要靠对酒店的套利机会赚大钱"。
**Lenny Rachitsky:** 他们只是需要付房租。
**Howie Liu:** 对,而且我觉得他们热爱这个产品,他们也热爱自己打造产品的方式,热爱那个产品和公司与文化中那种以设计为中心的特质。正是这些,让你在可能长达非常长的时间里持续经营同一家公司的过程中,依然能从中获得快乐。
**Lenny Rachitsky:** Howie,在我们进入非常令人期待的闪电问答之前,还有什么你想聊的或者想留给听众的吗?
### 每个人都可以成为全能型 builder
**Howie Liu:** 我想重申一下,尤其是对于这里在 EPD——特别是 P(产品)角色上的听众——我真心认为这并不是一个"你要么有要么就没有"的技能组合问题,在 AI 时代保持相关性的能力也是如此。但我确实认为这是一个行动号召——去强化那些你现在可能还不够精进的技能。甚至编程方面,我真心相信,如果愿意的话,每个人都可以学会做一名软件工程师。当然,显然有些人——就像优秀的写作者中也不是每个人都能成为出版作家或海明威——但每个人都可以获得足够好的软件工程熟练度,只要他们真心想要的话。你可以去上个训练营,可以做一些编程练习等等。我的意思是,有时候我们把这些学科当作硬性技能来对待——如果你已经到了职业生涯的中途,你还没有成为工程师,你还没有成为设计师,好吧,那你永远也不可能成为其中一个了。我只是认为我们的大脑具有可塑性,外面有大量优秀的课程可以学习。很多东西,就像我说的,归根结底也在于试错和做项目,也许是利用晚上和周末的项目来学习这些东西。在 AI 原生时代,每个人都可以学会成为一个多面手——那种产品工程师/设计师的混合型角色。唯一阻止你的,只是你没有走出去动手去做。
**Lenny Rachitsky:** 这是一个非常令人振奋的结尾方式。我想在此基础上再加一点——现在学习这些东西比以往任何时候都容易。你可以和超级智能对话,它们在你构建的过程中可以帮你学习。
### 学习技术最好的时代
**Howie Liu:** 是的。说真的,我有时候会打开 ChatGPT,直接问它:"嘿,这个应用你会怎么搭建?"我就是好奇。我会问:"你会怎么搭建 Manus,那个开放式 agent?"直接问怎么搭建。你可以提出这些问题,就像身边有一位了不起的、才华横溢的软件架构师、软件工程师、产品经理、设计专家导师——你可以毫无顾忌地提问,不存在愚蠢的问题。它有无限的耐心,全天候 24 小时在线。正如你所说,这是学习这些东西最不可思议的时代。然后当然还有那些可以让你实际动手构建东西的交互式工具。任何人都可以下载 Cursor,直接让 Composer 帮你生成一些代码,然后去看代码,试着弄明白它做了什么。正如你所说,回想我最早经历的构建应用的时代——我先学了 C++,然后学了 PHP 和 JavaScript,甚至在早期零八到一零年间构建 JavaScript 单页应用——那是一门非常黑暗的技艺。我是说,确实有一些资源……但你只能自己去摸索学习这些东西。没有很好的教程。你得逆向工程某些东西。还有些奇怪的事情——比如你想在 UI 里做圆角,你得打开 Photoshop,用像素画一个圆角,然后把那个像素切成一张图片,精准地放到页面上合适的位置,让它在盒子的边缘显示。
那时候都是这种疯狂的东西。一切都比如今要晦涩得多,而现在感觉流畅得多、容易接近得多,你需要涉猎的那些晦涩技术与你要构建的东西之间的鸿沟已经被极大地缩小了。你所付出的努力、你需要跨越的抽象层级,与你真正想要构建的那个神奇而令人愉悦的东西之间的距离已经被极大地缩短了。对于 builder 来说,这是前所未有的激动人心的时代。
**Lenny Rachitsky:** 你还记得 spacer.gif 吗?
**Howie Liu:** 哦,记得记得。
**Lenny Rachitsky:** 就是为了创建……就是那种线条之类的东西,你不得不——
**Howie Liu:** 对,我记得。是的。
**Lenny Rachitsky:** ……那个不可见的一个像素的东西,你到处塞到各个位置。
**Howie Liu:** 对。对对对。没错了。
**Lenny Rachitsky:** 天哪,活在那样的时代真是……Howie,说到这里,我们已经到了非常令人期待的闪电问答环节。我有五个问题给你。准备好了吗?
**Howie Liu:** 准备好了。
### 闪电问答
**Lenny Rachitsky:** 开始了。有两三本书是你发现自己最常推荐给别人的?
**Howie Liu:** 我最近一直在尝试多读一些小说,部分原因是我觉得这真的是一种很好的思维重置。我推荐《三体》给所有还没读过的人,这是一本拓展思维的书。我喜欢那种能打开你大脑的科幻和虚构作品,所以也许这是我在作弊,但它是一个三部曲系列,那就是三本好书。
**Lenny Rachitsky:** 我很喜欢那个系列,我的建议是读到一本半之后才开始精彩,所以坚持读下去就好。就是那种感觉:"好吧,现在我入坑了。"
**Howie Liu:** 我连第一部就很喜欢,而且我觉得它像《盗梦空间》一样,每一部续作都像是你又坠入了另一层,就像一层层深入下去,对吧?
**Lenny Rachitsky:** 太棒了。好的,你最近最喜欢的电影或电视剧是什么?
**Howie Liu:** 电视剧的话,我刚开始看《The Studio》。就是 Seth Rogen 主演的那部。
**Lenny Rachitsky:** 对,看起来特别焦虑。
**Howie Liu:** 是的,确实挺让人紧张的。而且说起来,当年《Silicon Valley》播出的时候就太贴近现实了,我看了,但真的尴尬到不行。《The Studio》看起来就挺有趣的,因为它讲的是好莱坞的内部故事,而我并不在好莱坞,所以看得挺开心的。我觉得这部剧很聪明也很有趣。因为我往返于洛杉矶和旧金山之间生活,所以也觉得特别真实。剧里塑造的那些角色,我在现实世界中真的能见到原型。
### 最近发现的好产品
**Lenny Rachitsky:** 你最近有没有发现什么特别喜欢的产品?可以是应用、小工具,也可以是服装。
**Howie Liu:** 好吧,我给两个答案,因为我觉得必须得说一个软件产品。我真的是 Runway 的超级粉丝,无论是产品还是公司。他们推出的每一个新模型都让我惊叹,最近两天刚出了一个新模型,对创建你想要的视频场景提供了更多的控制和精细调节。我觉得现在能生成的画面真实度非常高,而且他们还做了一个很酷的沉浸式世界生成器的 demo,我之前提到过。能亲眼看到这些真的很酷。我也喜欢这个以弱胜强的故事。显然 Google 正在大举进攻这个领域,有 VO3 等等产品,OpenAI 也在发力,但我喜欢这个不到百人的小公司仍在以小搏大、打造真正出色的视频体验的故事。这就是软件方面的答案。
然后,一个非常非常宅的实体产品答案是,我最近开始迷上了一个小众领域——日本小型制造商以匠人方式生产的服装,他们用的真的是有百年历史的织布机,以老式方法或老式工业方法制作衣服。他们有一种叫 loop wheeler 的机器,以非常慢的速度纺织布料,从量产角度来看完全不实际,但我买了几件这样的 T 恤,我就是喜欢那种……我觉得在这个一切都在加速变化、五年前的技术就已经过时的世界里,我反而喜欢这种回望旧物的感觉——有时候老东西在这个新时代反而更值得珍惜。也许这让我成了一个hipster,但我越来越喜欢复古的东西了。
**Lenny Rachitsky:** 我觉得凡是以"匠人手作小批量日本"开头的东西,肯定都差不了。有没有想分享的品牌?还是你想保持低调——
**Howie Liu:** 其实有一个叫 Self Edge 的,它在旧金山 Valencia Street 上有一家实体店,是他们的主店。他们经销很多这类商品,这基本上就是他们的经营理念,有牛仔裤、T 恤之类的。我买了很多。他们 basically 策划了一个非常精选的不同制作品牌集合。其中一个叫 Studio D'Artisan,还有一个叫……其实挺酷的,有一家公司……我觉得母公司就是 Toyo,T-O-Y-O,Manufacturing,听起来像是那种大规模的企业集团,但其实完全不是。它是一个规模很小的日本复古服装制造商,旗下有几个子品牌。
他们其实买下了一个美国战后品牌的版权,这个品牌当年有点像 Hanes,是当时男士内衣和运动装的四大或五大品牌之一,叫 Whitesville。我不知道这个名字的由来,但 basically 就是一些基础款服装,T 恤之类的。这家日本独立公司买下了这个已经停用的品牌名,现在几乎按照完全相同的版型、面料,甚至连 T 恤上的图案包装都一模一样地复刻出来,但是在当下生产。我觉得这里有一种非常有趣和反讽的东西——他们把美国战后的美学和品牌拿过来,但实际上是以日本独立小规模制造的方式在做这些衣服。
**Lenny Rachitsky:** 我感觉我们刚刚打开了一个完全可以做成另一整期播客的关于服装和——
**Howie Liu:** 对——
**Lenny Rachitsky:** 工艺的话题——
**Howie Liu:** 下一个播客系列。
**Lenny Rachitsky:** 或者就叫"Howie 和 Lenny 聊服装"。
**Howie Liu:** 挺好的。
### 人生格言
**Lenny Rachitsky:** 好的,还有最后两个问题。你有没有一个人生格言,在工作中经常觉得有用,或者喜欢和朋友家人分享的?
**Howie Liu:** 我偶然发现了 Paul Conti 这个人,他是一位医学博士,同时也是心理学家。他写了一本书,还在 Andrew Huberman 的播客上做了一期长访谈。他其实谈了很多关于如何看待人生观、如何建立思考人生的框架,但都是建立在科学、神经科学和认知科学的基础上的。我觉得有一个观点特别特别有力,让我铭记在心,那就是:如果你以谦逊和感恩为基础来经营你的人生。当然,每个人的处境不同。
我完全承认,虽然我家境不富裕,小时候家里经济条件非常一般,但我仍然获得了令人难以置信的资源和机会,仅仅因为我在美国长大,出生和成长在美国,而且能接触到电脑和互联网,以及由此获得的所有免费资源去学习和了解。但我仍然觉得,无论你最初拥有什么或不拥有什么,如果你以一种谦逊和感恩的精神去面对世界和未来,而不是走向反面,我觉得这会变成一种自我实现的预言。你心态开放,心怀感恩,然后更多的机会真的会来到你身边,也许是因为你向这个世界和其他人释放出的能量。
你在吸引好的机会、好的人和好的事物。他的框架中还有很多其他部分,但最容易记住的就是:我如何面对每一天?即使我正在经历一段困难时期,今天不得不解雇某个人,或者因为丢了一个客户订单而失望,或者出了什么故障,不管是什么,但仍然努力以整体的谦逊和感恩之情来看待整个局面——我觉得这真的会改变你的……它会渗透到那一天的方方面面,甚至可能影响你的一生。
**Lenny Rachitsky:** 这让我深有共鸣。这是一个非常有力量但很难内化的建议。
**Howie Liu:** 是的,说出来容易,做到很难。
---
**Lenny Rachitsky:** 大家可以在哪里找到你?关于 Airtable 有什么需要了解的?听众怎样能帮到你?
**Howie Liu:** 好的,我在 Twitter 上,账号是 howietl。我不怎么发帖,但我是潜水党,所以我会看会听,你随时可以在那里私信我。你也可以直接给我发邮件,howie@airtable.com,有任何想法、反馈等等都可以。关于 Airtable,直接去试试就好。我们的核心理念就是让这成为一个靠体验驱动的产品。这就是为什么我们非常坚定地回归 PLG 根基。我们之前聊到过,首页上直接写着:"现在就开始构建。你想构建什么?直接动手。" 点进去就开始搭建,所以去用产品,给我反馈。如果你自己有想法想要一起探讨碰撞,我非常欢迎,因为我的热情就在于思考产品和产品用户体验,尤其是在 AI 时代,如果你正在做或者在思考这个领域里有意思的东西。哪怕纯粹只是想就某个概念进行头脑风暴,这也是我很享受的事情,而且也许我能从中学习、磨砺自己的技能。随时联系我。当然,也告诉你的朋友和家人试试 Airtable,这是最重要的事情。
**Lenny Rachitsky:** 听起来你在等别人用有趣的问题来吸引你的注意力——
**Howie Liu:** 没错,是的。
**Lenny Rachitsky:** Howie,非常感谢你来参加节目。
**Howie Liu:** 太棒了。谢谢你,Lenny。
**Lenny Rachitsky:** 大家再见。非常感谢收听。如果你觉得这期节目有价值,可以在 Apple Podcasts、Spotify 或你喜欢的播客应用上订阅。也请考虑给我们评分或留下评价,这真的能帮助更多听众找到这个播客。你可以在 lennyspodcast.com 找到所有往期节目或了解更多关于这个节目的信息。下期再见。
## 术语表
| 原文 | 中文 |
|------|------|
| A/B test | A/B 测试(对照实验方法) |
| AE | AE(Account Executive,客户经理) |
| affordances | 可供性(交互设计术语,指界面元素暗示的操作可能性) |
| agent | agent(智能代理,保留原文) |
| Airtable | Airtable(产品名,保留原文) |
| All In | All In(播客名,保留原文) |
| alpha | alpha(指具有前瞻价值的信息/洞察,保留原文) |
| Andrew Huberman | Andrew Huberman(人名,保留原文) |
| APM | APM(Associate Product Manager,产品经理培训生项目,保留原文) |
| Brian Chesky | Brian Chesky(Airbnb 联合创始人兼 CEO,人名保留原文) |
| CB Insights | CB Insights(公司名,保留原文) |
| ChatGPT | ChatGPT(产品名,保留原文) |
| chief of staff | 幕僚长 |
| Composer | Composer(Cursor 的功能,保留原文) |
| CRM | CRM(Customer Relationship Management,客户关系管理系统,保留原文) |
| Cursor | Cursor(AI 代码编辑器,保留原文) |
| Dan Shipper | Dan Shipper(人名,保留原文) |
| decacorn | decacorn(百亿美元级独角兽) |
| Double Diamond | Double Diamond(双钻石设计框架,英国设计委员会提出的设计方法论) |
| DRI | DRI(Directly Responsible Individual,直接负责人制度,保留原文) |
| DSL | DSL(Domain-Specific Language,领域特定语言,保留原文) |
| Eames chair | Eames 椅(Charles & Ray Eames 设计的经典座椅) |
| EPD | EPD(Engineering/Product/Design,工程/产品/设计部门,保留原文) |
| evals | evals(评估/评测,AI 模型输出质量的系统化评估方法,保留原文) |
| Every | Every(公司名/品牌名,保留原文) |
| fast thinking / slow thinking | 快思考 / 慢思考(源自 Daniel Kahneman 的《思考,快与慢》) |
| field agents | field agents(Airtable 中在应用内代替客户运行的 agent,保留原文) |
| Flexport | Flexport(公司名,保留原文) |
| founder mode | founder mode(创始人模式,指创始人深度介入运营的管理方式) |
| Framer | Framer(原型设计工具,保留原文) |
| freemium | 免费增值(免费基础版+付费高级版的商业模式) |
| GitHub Copilot | GitHub Copilot(产品名,保留原文) |
| GM | GM(General Manager,总经理制,保留原文) |
| golden pathy | 理想路径的(指测试中的最佳场景路径) |
| GPT-5 | GPT-5(OpenAI 模型,保留原文) |
| growth mindset | 成长型思维 |
| Harvey | Harvey(AI 法律科技公司,保留原文) |
| HeyGen | HeyGen(AI 虚拟人视频生成平台,保留原文) |
| hipster | hipster(指追求小众品味的人,保留原文) |
| HyperDB | HyperDB(Airtable 的数据存储引擎,保留原文) |
| IC CEO | IC CEO(个人贡献者型 CEO) |
| jobs to be done | jobs to be done(待完成的任务理论,保留原文) |
| loop wheeler | loop wheeler(一种低速圆筒编织机,保留原文) |
| Lovable | Lovable(AI 应用构建平台,保留原文) |
| map reduce | map reduce(分布式计算范式,保留原文) |
| nerd snipe | 用有趣的问题吸引某人的注意力(网络用语,指用一个引人入胜的技术/智识问题让人无法抗拒地去思考) |
| Nick Turley | Nick Turley(人名,保留原文) |
| Nikita | Nikita(人名,指 Nikita Bier,保留原文) |
| no code | 无代码 |
| NotebookLM | NotebookLM(Google 的 AI 笔记工具,保留原文) |
| Omni | Omni(Airtable 的对话代理,保留原文) |
| OpenAI | OpenAI(公司名,保留原文) |
| Palantir | Palantir(公司名,保留原文) |
| Paul Conti | Paul Conti(人名,保留原文) |
| PLG | PLG(Product-Led Growth,产品驱动增长,保留原文) |
| PMM | PMM(Product Marketing Manager,产品营销经理) |
| POC | POC(Proof of Concept,概念验证,保留原文) |
| PRD | PRD(Product Requirements Document,产品需求文档,保留原文) |
| product-market fit | product-market fit(产品市场契合度,首次出现保留原文) |
| Replit | Replit(在线编程平台,保留原文) |
| Revolut | Revolut(数字银行/金融科技公司,保留原文) |
| Rippling | Rippling(公司名,保留原文) |
| Runway | Runway(AI 视频生成平台,保留原文) |
| SE | SE(Sales Engineer,销售工程师) |
| Self Edge | Self Edge(品牌名,保留原文) |
| Sesame AI | Sesame AI(AI 语音技术公司,保留原文) |
| Seth Rogen | Seth Rogen(人名,保留原文) |
| Studio D'Artisan | Studio D'Artisan(品牌名,保留原文) |
| v0 | v0(Vercel 的 AI 代码生成产品,保留原文) |
| vibe coding | vibe coding(凭感觉用 AI 写代码的方式,保留原文) |
| vibes | vibes(凭直觉感受,与 evals 的系统性测试相对,保留原文) |
| Whitesville | Whitesville(品牌名,保留原文) |
| Windsurf | Windsurf(AI 代码编辑器,保留原文) |
| YC | YC(Y Combinator,知名创业孵化器,保留原文) |
---
*此文档由 AI 分片翻译(translate_long_document)*