为什么AI正在颠覆传统产品管理 | Tomer Cohen (LinkedIn CPO)
Why AI is disrupting traditional product management | Tomer Cohen (LinkedIn CPO)
Tomer Cohen: When we look at the skills required to do your job, by 2030, it will change by 70%. So whether or not you’re looking to change your job, your job is changing. In order to stay competitive, you actually have to go back to some first principles, go back to the drawing board and reimagine what it means to be building.
Why Product Building is Changing
Lenny Rachitsky: You’re experimenting with a very different way of building product at LinkedIn that fully embraces what AI unlocks.
Tomer Cohen: We call it the full stack builder model. The goal itself is to empower great builders to take their idea and to take it to market, regardless of their role and the stack and which team they’re on. It’s really fluid interaction between human and machine.
Full-Stack Builder Project
Lenny Rachitsky: This feels like this could be a model for how a lot of companies operate and how product ends up being built in the future.
Tomer Cohen: Change management here is going to be a critical part, but it’s not enough to give them the tools. You have to build the incentives programs, the motivation, the examples to how you do it. I see a lot of companies roll out their agents and just expecting companies to adopt. It doesn’t work this way.
Vision for Full-Stack Builders
Lenny Rachitsky: There’s always been this question, is AI going to just make people that are not amazing, more amazing, or is it going to make amazing people even more amazing?
Three Pillars of Automation
Tomer Cohen: Top talent has this tendency of continuously trying to get better at their craft. The key trait that I’m emphasizing for builders is…
Lenny Rachitsky: Today, my guest is Tomer Cohen, longtime chief product officer at LinkedIn, who is piloting a new way of building that I think will become a model for how companies operate in the future. It’s called the Full Stack Builder Program, and essentially the idea is to enable anyone, no matter their function, to take products from idea to launch. They’ve scrapped their APM program and replaced it with an associate full stack builder program. They’ve introduced a new career path with the title Full Stack Builder that anyone from any function can become. And as you’ll hear in the conversation, they’ve built a bunch of internal tools and agents and processes to basically build a human plus AI product team that can move really fast, adjust to change quickly, and do a lot more with a lot less. If you’re looking for inspiration for how to rethink how your team operates and to lean into what AI is unlocking for teams and companies, this episode is for you.
A huge thank you to Shira Gasarch for suggesting topics for this conversation. If you enjoy this podcast, don’t forget to subscribe and follow it in your favorite podcasting app or YouTube. It helps tremendously. And if you become an annual subscriber of my newsletter, you get a year free of a bunch of incredible products, including a year free of Devin, Lovable, Replit, Bolt, n8n, Linear, Superhuman, Descript, [inaudible 00:02:29], Gamma, Perplexity, Warp, Granola, Magic Patterns, Raycast, ChatPRD, Mobbin and Stripe Atlas. Head on over to lennysnewsletter.com and click product pass. With that, I bring you Tomer Cohen after a short word from our sponsors.
My podcast guests and I love talking about craft and taste and agency and product market fit. You know what we don’t love talking about? SOC 2. That’s where Vanta comes in. Vanta helps companies of all sizes get compliant fast and stay that way with industry-leading AI, automation, and continuous monitoring. Whether you’re a startup tackling your first SOC 2 or ISO 27001 or an enterprise managing vendor risk, Vanta’s trust management platform makes it quicker, easier, and more scalable. Vanta also helps you complete security questionnaires up to five times faster so that you can win bigger deals sooner. The result? According to a recent IDC study, Vanta customers slashed over 1,000 off at vanta.com/lenny.
But Figma never felt like it was for me. It was great for giving feedback and designs, but as a builder, I wanted to make stuff. That’s why Figma built Figma Make. With just a few prompts, you can make any idea or design into a fully functional prototype or app that anyone can iterate on and validate with customers. Figma Make is a different kind of vibe coding tool. Because it’s all in Figma, you can use your team’s existing design building blocks, making it easy to create outputs that look good and feel real and are connected to how your team builds. Stop spending so much time telling people about your product vision, and instead show it to them. Make code-backed prototypes and apps fast with Figma Make. Check it out at figma.com/lenny.
Tomer, thank you so much for being here and welcome to the podcast.
Tools and AI Agents
Tomer Cohen: Thank you. It’s great to be back.
Lenny Rachitsky: It’s great to have you back. I’m really excited to be chatting because you’re experimenting with a very different way of building product at LinkedIn that fully embraces what AI unlocks, kind of leans into what is now possible, and to me, this feels like this could be a model for how a lot of companies operate and how product ends up being built in the future. There’s a lot of product leaders that are talking about AI, what they can do. It feels like you’re actually doing this in a really, really radical way, and so I’m excited to learn from you to hear about this for listeners to understand what you’re seeing, what you’ve learned. Let me start with just why did you decide this was necessary? Why are you rethinking all of these things about how product has been built for a long time? AKA, why do people need to pay attention to what we’re about to be talking about?
Growth Agents and Research Agents
Tomer Cohen: It really starts with kind of the basics. For me, technology has always been about empowerment. It’s not about what it does for us. It’s about what enables us to do. And now we have this amazing opportunity in my mind to make it about meritocracy, and I think it’s an opportunity, but it’s also a necessity right now, and I want to put this in context where we’re entering this phase where the time constant of change is far greater than the time constant of response. Basically means that change is happening faster than we’re able to respond to it. Now, LinkedIn has this unique view of the world of work. So we actually have some pretty, in my mind, mind-blowing stats to put this in perspective. When we look at the skills required to do your job, by 2030, which is literally four years from now, sounds a long time, but four years from now, it will change by 70%.
So whether or not you’re looking to change your job, your job is changing. The only question is, do you keep it? And then we look at organizationally, the fastest growing jobs right now, the most in demand jobs in the market are growing by north of 70% from last year’s fastest growing job. So there’s a new kind of iteration of what you need as an organization to thrive. And then you apply that to building products and you realize that in order to stay competitive, you actually have to go back to some first principles, go back to the drawing board and reimagine what it means to be building. And what I love about this is when you think about the role of a builder, which the builder is at the heart of company, the goal is actually quite simple. The builder takes an ADN, she brings it to life. That’s really the process, right?
And we all build those, let’s call them best practices. You research the problem really well, you spec it out, you design it, you code it, you launch it, and you iterate. That’s basically it. But what happens at many at scale companies, LinkedIn included and many other companies, over time that process became very complex very quickly. So what happened? We took every step and we expanded it to a lot of sub-steps. Researching, the problem became looking at for us 10 to 15 sources of information, obviously talking to customers about doing data pools, looking at feedback tickets in multiple sources, social media, interactions with customers. We probably have 10 to 15 sources of information we go for before we feel like we have research department really, really well.
Think about reviews for product. There is design reviews, privacy reviews, security reviews. I can go on and on and on. And each one of those substeps actually has a valid reason to exist. But when you add a whole thing together, you’re like, “Oh my God. This is why it takes, to build a small feature, multiple teams, multiple code bases, multiple sprints just to get it out to launch,” and not talk about iterating, which is actually where you seek success. You never see success in the launch itself. So really the work itself is not complex, but the process we made very complex. And when I was digging in, I found it doesn’t end there because somebody has to do all those substeps, so what happened is you actually move from process complexity to organizational complexity as well.
And then you actually led to microspecialization. All those subsets are doing by somebody specific. So from one builder, we have multiple functions. Obviously we have engineering, product and design, and you can start questioning those lines. At least I am internally. And from there, we have a lot of subspecialties. It happens in every one of those functions, but imagine design. We have interaction design, animation design, content design, research. There’s so many aspects to that. So they’re all valid, but they all have people, and that entire process basically means a lot of… It’s basically bloating. It’s complexity. And then without noticing, you end up with this massively complex… We actually have this diagram that basically shows the process complexity, organizational complexity together.
And usually people are mind blown because they’re working on one thing very specific, but when you zoom out, you have this overwhelming experience you’re kind of thinking about. And now we have this real opportunity to collapse the stack backup, go back to craftsmanship, rethink the product development lifecycle, which is where the full stack builder model comes to life.
Lenny Rachitsky: Wow. Okay. And there’s so much here. We’re going to be showing the visuals as you talk to help people see what you’re explaining here. And all of this is very rational. If you have 15 sources of information, why not pull from it? Why miss out on that stuff? And what you’re describing here is as you get more power and more specialized… It all makes sense rationally, but when you start to step back and look at this like, holy shit, it takes six months to launch one feature. I want to ask about the stat you shared. I think this is an incredibly powerful stat and you have very unique data here to tell you this sort of stuff. So you said that something like 70% of the skills that people will need in the future are going to change.
Agent Building Platforms and Tools
Tomer Cohen: To do their current job.
Single Responsibility and Orchestration Layers
Lenny Rachitsky: To do their current job. And what is this looking at? Is this just based on historical data or how do you find that?
Tomer Cohen: Yeah. To be fair, there was always a change, right? So it was never about just keep the skills you have today, but we’ve never seen such a dramatic part of your role today. So whether you are a marketer right now or a seller, a recruiter, an engineer. Engineering is where a lot of the investment is going in right now in terms of agents. Those jobs will change dramatically. I remember I said my role, my life as an engineer and even then it’s changed materially after 10 years, and then the change we’re seeing right now, just thinking about in four years, what did it take to actually engineer really, really well would be dramatically different, or to build software, to build an artifact of some sort. But it’s true for almost every function. It’s not equal. Some job like nurses will see less impact, but some jobs will see 90%, 95% impact.
Investment Focus and Coding Acceleration
Lenny Rachitsky: There’s also a stat that I don’t think you mentioned here that I saw on the post when you first talked about this program is that 70% of today’s fastest growing jobs were not even on the list of jobs a year ago.
Tomer Cohen: Yeah. No, so this is the fastest growing job on the list were not there a year ago, and then many of them don’t even exist a decade or two ago. There’s actually some pretty amazing stats across the board.
Current Pilots and Scale
Lenny Rachitsky: Okay. So let’s talk about this program that you built. Tell us the name and then tell us the gist of what it is today and the vision of where you want it to be.
Tomer Cohen: Yeah. So we call it the full stack builder model. And the goal, always start with the goal. The goal itself is to empower great builders to take their idea and to take it to market, regardless of their role and the stack and specifically which team they’re on. And the idea ultimately is to be able for that builder is to develop experiences end to end, to combine skills and expertise across what was traditionally distinct domains to bring it all together. And it’s not a sequence of steps. It’s really a fluid interaction between human and machine. That’s how the way I see it. And then when you look back at that product development life cycle from the idea, the insight all the way to launch, the key trait that I’m emphasizing for builders is where I want them to spend their time is where I think great builders should shine in.
So the idea of vision. Coming up with a compelling sense about the future. Empathy, super critical, right? Having a profound understanding of an unmet need. Communication is critical. And we see this a lot in job descriptions right now for almost every role, but ability for you to align and rally others around an idea. Creativity, which for me is about coming up with possibilities beyond the obvious. For example, I don’t think AI yet is great at creativity. I think it’s kind of, in many ways, brings back the things you might not know about, but it’s not the kind of next level creativity, which I think still humans are much better at.
And then ultimately what I think is the most important trait for a builder is judgment. Some people call it test making, but it’s making high quality decisions in what is complex ambiguous situations. Everything else, I’m working really hard to automate. Really, really hard. And then when you think about the outcome, it’s not just about having more shots at the goal, which I think people go like, “Oh, the iteration speed is going to be very high.” Yes, but what you’re really doing to an organization of at scale organizations is they’re a lot more nimble, a lot more adaptive, a lot more resilient. They can navigate the future. They can actually match the pace of change to the pace of response.
And an analogy I have in mind is kind of Navy SEALs. You come to training, they’re all kind of learning, they’re cross-trained, across multiple areas. What they specialize in is the mission and they operate in small pods and they’re very nimble and you can assemble them very quickly. And I think that’s going to be the organization that will win in the future.
Pilot Rollout Strategy
Lenny Rachitsky: Okay. So the simple idea, if you’re just to boil it down to a sentence, the idea here is there’s a builder who goes through the entire product development process essentially on their own. They have an idea, they research, they do data, they prototype design ship. That’s kind of like the vision of where this goes?
Tomer Cohen: Yes, but it doesn’t have to be on their own. It’s not like… I still believe in teams.
Entry Projects to Full Experiences
Lenny Rachitsky: Got it. So smaller teams.
Tomer Cohen: Just smaller teams. Smaller teams and much more focused on the problem, the mission, per say, versus… Actually, one of the things we’ve done as an example, we started to do the idea of pods. We’re no longer large teams. We assemble a team, ideally a full stack builders coming together and it’s less about can I have an engineer design PM working together and trying to combine this trio looking at folks who can flex across and then they tackle something for a quarter or so and then we reassemble those two different pods. That’s one example of an manifestation we’re doing right now and seeing actually some great success in both in terms of velocity, but also in terms of that focus and nimbleness of that team.
Adoption by Top Talent
Lenny Rachitsky: And it feels like the goal here, what you’re trying to adjust and that broke as teams bloated as speed and adaptability and flexibility, because going back to your original point that change is happening so much more quickly now that companies that have been building in this traditional way just can’t compete.
Tomer Cohen: Yeah. It’s not that you have to break the model. I think the model is broken. It’s just this pace of change is helping us realize it.
Change Management and Culture
Lenny Rachitsky: Okay. So then going back to the things that these builders still do versus what you want to automate. So the list you shared is they’re responsible for the vision, empathy, communication, creativity, and judgment.
Tomer Cohen: Yes. Yeah. And I would put a lot of the focus on the latter. I think if you ask me at the end of the day, what’s the kind of most important trait? I would say it’s that judgment, test making ability.
Encouraging Autonomous Tool Exploration
Lenny Rachitsky: And then in terms of what you’re automating, what are some of the areas you’ve seen a lot of success in actually automating and where do you think this goes?
Tomer Cohen: Yeah. So I think just to kind of break it to pieces, and I think this is… If you were a startup right now, in many ways you can start this way. There’s no legacy code, there’s no legacy structure you run. And in fact, a lot of the startups I talked to that are built AI natively, they’re just working at full stack builders. That’s the way they start. If you’re at a company at a scale of ours and many others in the market, you’re like, this is almost like a new production function and mindset that you have to do. And there’s really three components that we’re working on. One is platform. The second one is the tools and the agents. And lastly is the culture.
The platform one, this is the kind of level of investment you have to do before, before this actually starts, you start to see all the benefits accrue. But the platform for us as an example is rearchitecting all of our core platforms so AI can reason over it. So we’re building kind of this composable UI components with server side that we actually build. We’re basically building for AI to be ready to bring it in. So you can’t just go and bring a third party tool and have it work on the LinkedIn stack. In fact, that’s one of our biggest learnings. It never works. Never works. You have to bring it in and customize a lot of it, working almost in alpha mode with those companies to make it work internally.
Unexpected Negative Outcomes
Lenny Rachitsky: So this is essentially re-architecting your code base to work more efficiently with AI. Is that one way to think about it?
Tomer Cohen: Yes. And in many ways, working with those companies to adjust something in their stack to work with our stack as well.
Career Paths for Full-Stack Builders
Lenny Rachitsky: When you say those companies, meaning the development agents like Cursors and [inaudible 00:18:16] and such?
Tomer Cohen: Yes. Or Figma on design. Or you can think about design systems is another example of that. But you have to have that back and forth because they’re not… In many ways, we haven’t seen anybody be able to work off the shelf immediately on our code-based design systems and unique context we have.
Motivations and Timing for Role Transitions
Lenny Rachitsky: Just to follow that thread briefly, so there’s Figma. That’s interesting. So basically the way Figma exports and keeps your design system, that has to change to work better with AI is what I’m hearing.
Tomer Cohen: They first need to know how to work with our design systems, which is something, in many ways a lot of those companies are working on. Same with coding. It doesn’t work that you just bring it in and it just reasons over your code base really well. We tried. We are building that layer that basically allows it to do so, whether it’s Copilot or Cursor, Windsurf and so on.
Practical Tips for Transformation
Lenny Rachitsky: Got it. Okay. Oh yeah, Copilot. Microsoft. I get it. I get it. Okay. Okay. So that’s the platform. So that’s an investment that you guys have to make to make AI effective at building and doing all these things.
Tomer Cohen: And then you have tools. So tools is where you really build the agents. I mentioned I want to automate everything outside of those five trades that we talked about, and then we’re building the tools for that. And then for that, actually very similarly, I can’t just bring a tool from the outside and work. So I’ll give you an example. One of our biggest things is building a trust agent. Trust is really important for us at LinkedIn. There’s a lot of unique vectors which trust plays at LinkedIn doesn’t place it anywhere else. So we need to bring all of that know how and context and information base into that agent. So we ended up building our own trust agent at LinkedIn.
Rapid Fire Lightning Round
Lenny Rachitsky: And so what is this trust agent doing? Telling you when you’re maybe exposing information that you shouldn’t be?
Tomer Cohen: So when you build a spec, you build an idea, you walk through the trust agent and it’ll basically tell you what are your vulnerabilities, what harm vectors potentially you’re introducing or will be introduced as a result of that. And I had our head of trust build it. So the head of craft for every area is building their own agent. As an example, we have one of our features for job seekers is called Open to Work. If you’re looking for a job, you can put an open to work.
Movie and Podcast Recommendations
Lenny Rachitsky: Yeah, a little green loading thing on the circle.
Tomer Cohen: Exactly. And actually it’s a great signal. I’ve seen some great success from it. People are helping each other. The community really thrives around helping each other. But at the same time, it introduces a trust vector for bad actors because they’re open to work. People who are looking for a job are potentially more vulnerable to scams than other folks. So being able to think about how do we prevent all of those ahead of time. So we walked that spec from a couple of years ago through the trust agent. Not only was it able to find all the stuff we initiated at the beginning, but all the holes that we did not catch until later. So that’s a great example of something that actually worked really well.
That’s one. The other one is a growth agent, as an example. Again, LinkedIn has a very unique… Actually, we have an incredible growth team, growth process. We’ve kind of funneled all of our unique loops, our funnels, our tests of the past, everything into this growth agent, and now you can basically rock your respect for it, your idea for it. And it would not just allow you to do it better. It would actually critique how good is your idea. This is something you cannot bring off the shelf. It’s very unique to LinkedIn. So we had to invest dramatically in it. And one team which is using it right now, which is almost… I wasn’t thinking about it at the beginning, but our UXR team, our UER team, the user research team is usually using that growth agent to understand out of all the things that are basically surfacing for members, which one has the biggest growth opportunity to have the biggest impact? That was not in the cards when we thought about that idea, but teams are basically funneling those ideas into this one.
An example is our research agent. So research agent basically is trained on the personas of our members. You can think about a small business owner, a job seeker and so on. And it’s using not just world knowledge, it’s using all the research we’ve done in the past, all the support tickets coming in. So it’s pretty good at understanding that persona at LinkedIn. So one examples we had is a team came out with a spec. They weren’t aware we had the research agent yet. I asked the research agent for a small business owner, wanted to think about the marketing spec we had, and it critiqued it extremely well. Actually, in many ways shifted the direction of the team to focus on other integrations tools we can focus on, but it’s very hard to have that visibility all to all that corpus of knowledge inside of the company.
That’s another example. We have an analyst agent trained on all how you basically can query the entire LinkedIn graph, which is enormous. And instead of relying on your SQL queries or data science teams, you can use the analyst agent. All of those I would say are, I would call them still MVP+. The goal for us in the next couple of months to basically roll them out externally. Externally, I mean, internally at LinkedIn.
My Current Favorite Products
Lenny Rachitsky: Not as new product lines.
Life Mottos and Growth Mindset
Tomer Cohen: Exactly.
Lenny Rachitsky: Okay. So many questions. One is just how are you building this? Is there a platform you’re using? What does it take to build an agent at LinkedIn? Is it all internal tools or is there third party use?
Leaving LinkedIn and Future Plans
Tomer Cohen: It’s a great call. So I think we’ve been experimenting with a lot of tools. And I would say for a lot of those kind of knowledge corpus agents, we’re using everything from Copilot Enterprise to ChatGPT Enterprise. By far though, the most important part was basically our own customization of it. That’s been where we saw the biggest gains. Even building the orchestrator across those because you want the agents to start following to each other, the trust agent should work with the growth agent and go do a back and forth versus doing what more sequentially. So we’ve done a lot of work internally to make it happen. This is why I think it does require that level of investment.
And then in some cases, let’s talk about the design agent that we’re working with. We’re working with multiple companies to try and understand which product works best for us. And interestingly enough, and this is another learning, different teams gravitate to different products. So that’s something we’ll have to resolve and think about how we do this really well, because ultimately we were trying to simplify the process as much as possible, but that was a big learning for us and which tools we use and how we basically integrate them in.
Lenny Rachitsky: Got it. So you might have an amazing Figma agent, but some teams want to use a different design tool.
Tomer Cohen: Yeah. So we’ve kind of experimented with Figma and Subframe and Magic Patterns and so on, and we saw people gravitating depending on the function, their level of visibility, their know how of the tool before, they’re gravitating to different tools. And ultimately, I don’t want to have eight design agents in the company, so we have to converge into at least a few. And I think it’s similar across many areas because the appeal of those, a lot of those agents are trying to solve similar end goal, but they’re doing it very differently. And what you’ll see that ultimately, I don’t think there’s going to be a winner takes all because the starting point of the customer or the user will dictate a lot how simple they are for that use case.
Lenny Rachitsky: Super interesting. The other interesting takeaway here is you’re designing very specific agents that are one job to be done. Is that a very intentional decision? Did you try an agent that just is super intelligent on all these things?
Tomer Cohen: Ultimately, they will do an orchestrator. We’re going to really orchestrator across, but we did want to be able to rate and grade those agents really well on how they’re doing. And I think there is a level of expertise. Now, we’re kind of building this in a way where we’ll be able to mask a lot of those. You might not know that there’s a trust agent. You might have, we call this internally the product jammer agent that basically does your product jam, which is a process we do internally. You might just use the product jam engine, and that product jam agent will work with all the other agents. But now we’re starting with that building blocks until we build the orchestrating layer across.
Lenny Rachitsky: Another interesting takeaway from what you’ve been sharing is that so much of the work has gone into the beginning of the product development process, just like helping you craft the right requirements, clarify trust, and then here’s product jam and here’s the research we’ve done. And I imagine it’s because coding has already been accelerated with all these IEE tools. Talk about just why that’s maybe where most of the investment’s gone and where you’ve seen the most impact so far.
Tomer Cohen: Well, 100% our coding investment has gone, started a while back, and those are fall into place. We have our coding agent. In fact, we’ve kind of staged it into two parts of it. There is the idea to design part, and then let’s call it the code to launch part. The code to launch part has gotten a lot of attention and we’re making some big inroads there. Everything from the coding agent to what we call the maintenance agent when you have a failed build, it will do it for you. In fact, I think we’re close to 50% of all those builds being done by the maintenance agent and a QA agent.
Lenny Rachitsky: Wow. So this is when a break builds instead of engineers hopping on the issues that an agent fix.
Tomer Cohen: You can still go and finish your coffee before you have to go and redo the build again.
Lenny Rachitsky: Extremely cool.
Tomer Cohen: But we haven’t had much investment until we kind of launched this program in the idea to design area. And that’s a material part of work. It’s also where the quality a lot of the work comes from, at least before you start to go into the coding phase. The idea is to empower everybody. So if you’re an engineer, you can basically use all those tools at the front of the process and be able to be a full stack builder.
Lenny Rachitsky: How long did it take to get this kind of in place for you to actually form your first team to build these, the initial agents and some of this backend, redo the code base sort of thing?
Tomer Cohen: I announced this internally end of last year, we really kind of started working, but it was more setting up the teams and the processes internally. We had our first MVPs of those agents I think like four to five months after it was really trained, I would say. But really the work itself has been kind of couple of months of dedicated work. A lot of it has been getting all the corpus of data together, cleaning it up. And that’s actually a good learning as well. It’s not great to just give it access to your drive and say, “Reason all over this knowledge base.” It actually does a very poor job understanding importance of the past and putting weights on stuff. You actually want to think about specifically what the context when do you want to give it to and what’s the knowledge base that you want to have it focused on. So even cleaning up, let’s call them gold examples or golden examples to learn from, has been one of the biggest learnings. Just reasoning over your entire knowledge base did not work.
Lenny Rachitsky: Yeah, that makes sense. There may be just like a researcher with a strong opinion about something that you disagree with and it wouldn’t know. It’s like, oh, of course, this is data, this is fact.
Tomer Cohen: Exactly. And then it doesn’t always understand ties to original specs to success. You have to actually build… This is a really interesting way. When you think about how you bring those tools in, you can’t just bring them in. You have to know what you feed them with. And what you feed them with is not just access. I see a lot to just focus on the connectivity and integration and it reminds me of the… This is almost like, this is actually more than 10 years ago when I was co-rebuilding the team, co-rebuilding the feed at LinkedIn and we started from scratch and I had to literally sit down and filter through examples of what is a good professional post on LinkedIn and what is not. And this was like weeks of work getting up that golden sample of examples, but it wasn’t… The most important part was feeding at the right data, not all the data.
So it requires work. This is where I would say for many companies who are thinking about this phase, and I do a lot of sessions today with CPOs and COs on this process. You have to put this initial work to get the gains after. When I think about it, I think there’s a takeaway there in generally with AI, even if you’re learning it for the first time and so on, whether it’s Cursor or whether it’s design, if it’s Figma or other tools or Lovable, you should be ready to invest those hours before you start seeing yourself pick up in velocity and quality, which will come up, but you have to invest that time.
Lenny Rachitsky:
What’s the current state of the pilot? How large is it? How many teams are doing it? What kind of stuff have you shipped? Just give us a sense of today’s world.
Tomer Cohen: Yeah. I wouldn’t say we are yet at a very high sample rate where it’s kind of a high percentage of the organization, but we have a substantial part of the organization already using it to provide a lot of the feedback. We’re seeing a lot of great examples. So the way I think about the benefits is a function of experimentation volume multiplied by quality. How good are those experiments divided by the time it takes to actually pull them out, like idea to launch. So on saving times, we’re seeing, whether it’s PMs, designers, engineers, saving hours of work a week right now, whether it’s the analyst agent we talked about or the prototyping really quickly or the product jamming experience has been a big part of that. On the quality side, we’re seeing insights discussions just be much, much better. And by the way, quality and time, sometimes they help each other because it’s high quality, you don’t have to spend as much time on something.
So we are seeing that applied in. And the volume, I wouldn’t say we had a rate where I’m seeing a high percentage organization doing it yet, but this will come once we… We haven’t GA’d this internally. That will come in the next couple of months once we have all the stuff in place. But we’re seeing designers and PMs picking up bugs directly from Jira tickets, pushing them in, something we haven’t seen before, and there’s just an appetite for everybody to just join. So in fact, the biggest thing right now is everybody wants access. Everybody wants access to the tools to be able to do it together, and we just want to make sure it’s good enough to make sure the whole organization can do it really well.
Lenny Rachitsky: So how is it that you’re piling it? Is it some number of people have access to these agents and they just work the way they’ve worked with access to these tools? Or is there a team dedicated, this is the way you work now and this is it, and we’ll see what happens.
Tomer Cohen: So that’s a great call. So basically we have a team building. It’s the core team building the FSB track across all of R&D, FSB, full stack builder. And then there are pockets and pods of teams using it. So basically we are looking at specific areas that we’re basically giving it to. The condition there is they give feedback. As a response for that, they make the tool better, so it’s not just access. We want people who will use it. So one of your early adopters would be the ones who help [inaudible 00:34:15] up the product really well. So we’re doing this in a pod model right now.
Lenny Rachitsky: So it’s like a pod within a larger team, like a designer, PM, engineer kind of group within… Is there an example? You have a part of LinkedIn that’s trying this out?
Tomer Cohen: Yeah. So if I think about some of our teams, whether it’s… Actually, we just launched Semantic People Search and the Semantic Job Search as well. That team was using part of those tools to actually help build it. So that team actually, this was PMs building their own dashboards with those tools without waiting for design resources to come in. Then we have a design team who is now… This started really from the manager rolling this out. And in many ways, what I tell this team is, “Don’t wait for the official GA. Start doing it. Start leaning in.” We’re seeing designers of that team starting to push PRs, which never happened before. And now other teams, they want to do this as well. So it’s starting with this kind of grassroots experience. I would say the places have been very formal. I would say the beginning has been the top.
The product executive teams, basically we move from functional leaders, design, PM, BD, and so on to product areas leaders, and they basically rock across the stack and they also go for a 360 with all of those functions to see if they’re really able to do a full stack building experience. Then we’re also launching at the junior side a new program called the Associate Product Builder Program, where basically we used to have our APM program, which this is about it’s ending this year. And then starting January, we’re going to start having our APB program and they’re going to come into LinkedIn. We’re going to teach them how to code, design and PM at LinkedIn. They’re going to go through a pretty rigorous training process, and then they’re going to join those pods, and gradually we’re going to grow that program to be a material part of LinkedIn as well.
Lenny Rachitsky: Wow. So this might be the future of the APM program is this full stack builder APM-ish program.
Tomer Cohen: In many ways, we’ve built some pretty amazing… I’m really excited for that group. I wish I could join it. But we build amazing training for them. And in many ways, we’re going to use that training to think about how we roll it across the organization. We’re kind of using the lens of you have great technical skills, but you’re not an engineer at a company yet, or you have great design taste, but you haven’t designed at scale in company yet, and we’re going to teach you how to do it at LinkedIn, but the training we’re going to use a lot to extend across the company as well.
Lenny Rachitsky: Okay. So you have these programs, these pilots and these pods, and you said what you’re looking at to see if this is something you roll out is experiment velocity times quality times time.
Tomer Cohen: Divided by time.
Lenny Rachitsky: Divided by time. Okay.
Tomer Cohen: Yeah.
Lenny Rachitsky: Got it. And I guess I know it’s early, but just you said you’re seeing that it’s saving teams a few hours a week at this point, something like that?
Tomer Cohen: Yeah. And I think the feedback has been the most important part. Right? The way to think about this is just like you build a product. So we’re building this product internally and you want to experiment with some kind of early adopters who will give you feedback, and the feedback has been amazing. In fact, our top talent are the ones who are using this the most at LinkedIn. And the feedback from them has been incredible in terms because they’re also willing to spend the time and give the feedback as well. And the response from them has been incredible in terms of like the quality of their output, the time they’re spending on this to get the value back, their desire to be part of this and actually scale this and make this even better. So that’s where a lot of the excitement has been from how they’re using it and the quality we’ve seen there. I would say in six months or so, we’ll be able to see a lot more of the organization use it and you’ll start seeing those top line numbers will build as well.
Lenny Rachitsky: That is a really interesting insight that the top performers are finding the most success, because there’s always been this question, is AI going to just make people that are not amazing, more amazing, or is it going to make amazing people even more amazing? And it sounds like it’s likely the latter.
Tomer Cohen: Yes. And in many ways, it’s surprising, it’s not surprising. I’ve seen this also when we were… It’s surprising because you want everybody else to be part of this and lean in. I think top talent has this tendency of continuously trying to get better at their craft and this innate need to be at the cutting edge of how you build, and I think we’re seeing this here as well. This is why I had this phrase I say with the team that if we build all those tools, will they use it? And I know right now the answer is no. It’s not enough to give them the tools to use it. You have to build the incentives programs, the motivation, the examples to how you do it. They need to see other people being successful as well.
And I’ve seen this also when we’re shifting LinkedIn from a desktop company into a mobile company. It was a very similar process. It’s very hard. Change management here is going to be a critical part. I think I see a lot of companies roll out their agents and just expecting companies to adopt. It doesn’t work this way. Some will adopt. That tends to be your cutting edge 5% of talent that just wants new tools and they have a bias for change. But the vast majority needs to work for change management in how they do it, and that requires being a lot more thoughtful about the cultural aspect of it, which is by far from me the biggest and most important thing to do.
Lenny Rachitsky: Yeah. I want to spend time there. And it makes a lot of sense why people don’t spend time here because they have so much to do. They got to ship things. Their days are already busy. You have to now carve out time to learn this new tool that’ll not pay off for a while. So I get why people are like, “Okay, okay, I’ll get there. I’ll use it someday,” but they don’t. This idea of culture, when I saw you share this initially, this is the third piece of making this successful. So there’s the platform of getting the code base ready for people for AI to work with. Then there’s the tool, like the agents you’ve talked about, and then there’s the culture. Is there more there that you can share of just what has actually worked in helping get people on board? One thing I heard is creating a little bit of FOMO of like, okay, only a few people can use this and you have to sign up to get access. What’s worked in getting people to get on board?
Tomer Cohen: Yeah. I think this is where I emphasize to people that getting everything done, the platforms, the tools is not going to be sufficient. It’s a prerequisite for this to work, but not sufficient for this to work because it really requires you to invest a lot in the cultural aspects of how do you get people to lean into this one. And this one might feel slow at first, but I’ve seen this before with our transformation of thinking from desktop to mobile. And once it picks up, it actually maintains very high velocity. One, people are really incentivized by how you define expectations for them. So to think about what is the expectation of somebody in the role, whatever-
Lenny Rachitsky: So like changing performance review sort of things.
Tomer Cohen: Very much so. So everything from how you hire to calibration and evaluation. And one thing I want to see there early is this kind of AI agency and fluency. Like I mentioned, the tools are there. The question is, would you use them? Because the tools will be good enough, but not great at the beginning. That’s the classic thing of every good MVP tool. They’re good enough, but they’re not great. And then you kind of want to build that agency to make the tool better. We’re in this kind of notion of we’re going to make this better for LinkedIn together. Two is piloting success inside of your organization. That’s the pod model where you’re showing that not only this could work, it’s actually having success. So we have even our partnerships team, our BD team, being able to go instead of relying on waiting for an engineer to help build the developer portal and build the connectors there.
Literally one of our head of partnerships just went and did it himself. Didn’t even delegate to his team. And their goal is to say like, “Hey, I can do it. You can do it as well.” Those examples are really, really powerful. I talked about the associate product builder program where we are going to be very focused on training. I think that will send a really strong message across the organization. People will see this talent and what they can do, and I think that will create that movement. But celebrating wins in all hands, highlighting people and showing those examples. One example we’ve seen recently, people really looked at it in a surprise lens, but then it kind of, I think, really opened up a lens for them. We had somebody in our user research team. We had an opening for a PM on the growth team, and that role was open for a while, and she said that, “I think I can do it.”
And she used all these tools. This is a user researcher becoming a growth PM, not usually the career path you see, but she was excited about the area. She used all those tools, and she’s now a growth PM on the team. And really, you can start thinking about her more as a full stack builder ultimately. But seeing those openings and then highlighting those two people, actually people who are doing this have been a great example of it. And then just making sure that those tools are accessible. People can provide feedback, you share a lot, has been an incredible part of this. It’s not enough to be top-down directive that this is how we want to work. People want to feel like there are success stories. They feel like it’s worth their time. It feels it’s a movement they want to be part of, and then ultimately they can see successes in how they do it.
Lenny Rachitsky: I love this kind of comparison to the shift to mobile. We all went through that and there’s all these stories of companies requiring you to show mobile mocks. That’s the only way we’re going to operate. Now everything you have to ship has to be on mobile, and it’s interesting how similar this is to them, to that experience. And so a few things you just shared here just to kind of summarize some of the things that have worked for you. Showing wins, celebrating wins, showing people what other folks are doing with AI tools, creating a program that people enroll into and make it a little bit exclusive. This performance review piece is really interesting because that really will change people’s behaviors. Here’s how we get promoted. Have you actually already made that change to the PM? I guess it’s every track, I imagine, not just product management. Have you already made that change or is it kind of like a work in progress?
Tomer Cohen: So there was two aspects to it. Once I moved my team, my directs, we did 360 for them. So their 360 was, if you came from PM, you had the designers on your team rate you. And so that had its own, and then we shared those with them, and that had its own kind of motivation. But then we broadly took it across. So when we hire right now, we look for those. And then this upcoming cycle, we do a bi-annual. That’s going to be part of the performance evaluation piece and we announce it to everybody. And for what, it’s where people are excited to show. And they’re excited to know how they’re going to be… It’s always about, like, “I want to know how I’m being rated or evaluated.” So just being able to show those examples has been a big part of it.
The other thing I would say, it takes time for this program and its formality to roll out across the entire organization, and I was intentionally not trying to be quick at rolling this out to everybody because I think that just dilutes the value of it really quickly because it’s not about… I could care less about your title. I care about how you work. So calling you a full stack builder is not what I’m looking for. Changing your mindset to a full stack mindset is what I’m looking for. You’re thinking you can do the whole thing. You’re looking at those tools and looking at how to do it.
So one of the things I’ve said is if you’re looking for a formal reorg or declaration to start building differently, you’re waiting too long. Look, my biggest thing is here’s a permission for me to just not wait and just go. So whether or not you have the right tools or not, go build the tool, use a tool from the outside, bring it in, show those examples. In many ways, prove that you are a full stack builder in mindset before anything else come to mind. And that just naturally will happen, and that’s also where we’ve seen some of our best talent just goes and leans a lot into.
Lenny Rachitsky: I love that. I was going to actually mention that quote. Someone you shared, you work with told me exactly that quote you just shared, so I’m glad you brought it up of just if you’re waiting for a reorg, you’re not thinking about it the right way. How do you encourage people to actually play with these tools on their own? Are you just like, “Go take a few days to play with AI?” Is it just try it? Or is there anything formal you’ve seen of just getting people to more try this on their own without joining this program?
Tomer Cohen: A lot of the tools we’ve made, we’ve been sharing them regularly. A few of my all hands have been all about how to use those tools. But then at the same time, we’re kind of inviting, have you found a new tool that works really well for you? Share it, show it. Again, it could be Slack, could be Messages, Teams and so on, how you do it. But the idea is really to start getting that investment in how things work. Actually, I think in general, you can feel overwhelmed by tools right now, by recipes and how to do things like what’s your prompt and what’s my prompt. But really it’s finding something that kind of works really well, that can gravitate around and really invest in that’s been those areas. But I think we’ve had this invitation to go and explore and go and bring in stuff that you think are great. And in many ways, bring others along on the journey. It’s one good way to make the influence much bigger than a few folks who are doing really well with this.
Lenny Rachitsky: Are there any surprises on the negative side that have come out of this, of PRD is just feeling like AI driven, people slowing down unexpectedly? Is there anything that surprised you of just like, “Okay, this is actually not great”?
Tomer Cohen: Yeah, we mentioned a few of them. I was hoping for some tools to work off the shelf really well. It was never the case because we had to invest quite a lot.
Lenny Rachitsky: Never the case.
Tomer Cohen: Never the case. We had to invest quite a lot. And again, part of it is we just have a lot of legacy information and code based and knowledge and designs and so on. So a lot of the companies we work with are seeing this as a great growth opportunity for them as well to invest, but I do think it’s a big area of investment as well. We talked about not just giving access to all of your context which we started with, and we were like, “Oh, here’s access to all the drive, all information,” failed miserably and hallucinates like crazy.” People gravitating towards different tools, like our goal was to converge on tools, but that was pretty hard.
And then I think in terms of quality, we’ve just seen better quality, but I think it’s because, again, where we are in the stage is still the early adopters and they’re doing a few iterations in terms of how to do it. But I would say the tooling adoption is hard. And then I think for some people, this is important for me to kind of state, some people do not want to be full-stack builders, and that’s completely okay. Some people see themselves in specialization, and I think specialization has a place and a role. So I didn’t want the message to be across the organization I expect everybody to be a full-stack builder. I do not. I think there are system builders that empower full-stack builders, and then you have people who are specialized. But I don’t think we need as many specialized people as we did in the past.
Lenny Rachitsky: I didn’t actually realize this until just now. So is this their title now instead of product manager engineer, they’re full stack builder?
Tomer Cohen: We have a full stack builder title formally inside the organization, and we are gradually putting people in that bucket.
Lenny Rachitsky: So there’s a whole career ladder that’s forming. There’s a whole… Okay. That’s a bigger deal than I even thought. So where are you finding these folks mostly coming from, like product, engineering, design? I imagine it’s a mix, but just is there a most common trend?
Tomer Cohen: It’s a mix. People listening, I would just think about just go over your org and imagine who can do it, who can right now flex across those functions, whether it is engineering, design, product, even BD, and what you’ll find is there’s already quite a few that can flex across.
Lenny Rachitsky: Interesting. Are there any functions you think are especially successful at this? Not to play any favorites, but I don’t know. Are you finding like, okay? Or you could also not highlight any specific.
Tomer Cohen: No, I think it’s a mental model of how you do it. I think if I were to play what’s the hardest craft to potentially learn, I think design has a lot more work to get the design agents to be really, really good. So I think designers have a little bit of a leg up in terms of others learning their craft than the vice versa. But I honestly think it’s a mindset. I’ve seen designers code, I’ve seen PMs kind of design and do well. And this is why I think when you kind of step back and you think about people in your organization and who can flex, I think you’ll see them show up in many areas. And what I think you’ll find there is they have the agency, they’re leaning into new things, they have the fluency, like they’re already building new experiences and they have that growth mindset that they just want to get better, so it doesn’t matter what they learn at school or what label somebody put in them when they join the company.
Lenny Rachitsky: What I love about a lot of this is it’s the easiest time to transition between different product roles than it’s ever been. Design’s moving to PM, and sure, or just moving to this new role, it makes it so much easier to, like you said, that researcher became a growth PM.
Tomer Cohen: And this is probably my biggest advice slash motivation I give to the team because what I tell them is ultimately… By the way, this is for me as well. I think about it the same way. The incentives for you are so aligned with your organization of what we’re asking for, right? Because we need you to change. We want to be a more agile, adaptive, resilient organization that can deal with the pace of change, but you want as well for your own career. You want to be at the cutting edge of how you build. So the incentives are really aligned between what you need for your own career and what the organization needs you to do. So there’s that permission to go and do it for me is ideally kind of a tailwind in what they want to do more than anything else.
Lenny Rachitsky: Maybe a last question for people that are inspired and like, “Okay, this is what we need to be doing,” any just tips for someone starting down this road to be successful at trying something like this at their company?
Tomer Cohen: I would say I would start with the notion of how do you want to bring this just structure. I would think about the platform you need to build, the tools you want to bring, and then I would spend a lot of time on the culture. Platform and tools I think would be, again, a prerequisite, but not sufficient, and the cultural aspect is really important. I would think a lot about how you bring people along. So for one of the learnings we had that probably able to do it differently right now, if I were to redo this program was, for a while I was working very closely with my core team on it, the core kind of full stack building team that were in charge of building all this material, but the organization was always asking questions. “What’s going on? Who is doing it? What are the tools?” And in retrospect, we could have done a lot more in the flow to just show them and get them to already use early tools or be aware of it versus doing a small team on the side.
So it’s okay to start with a small team. I think it’s really important. But at the same time, just making sure there’s visibility across the whole thing is really powerful. Being patient and being willing to invest. I always give this example of, we always give this example of like, “Oh, look at this startup. They built this in a week.” Yes, you can build lifestyle in a week right now if you start from scratch. It’s actually not hard. But when you are trying to transform a large organization, you want to have this impatient about the goal and you have to have a high ambition, but being very thoughtful and patient about how you bring it to life and the key things you have to invest in. If you don’t invest in your platform, I just don’t see how this could be a successful outcome. If you don’t invest in customizing the tools for you, then you’re just going to get vanilla generic agents from the outside.
So being aware of the investment and making sure you actually allocate resource to it, this is kind of the classic, be willing to invest upfront so you can reap the benefit after, versus saying, “Hey, why am I not seeing us moving into 2X the productivity in a week?” That’s not going to be this way. You can see it with some people, but starting to collect those examples and starting to really think about the transformation is really key.
Lenny Rachitsky: This is so incredibly cool. I know that a lot of CPOs and heads of product and all kinds of leaders are reaching out to you trying to figure out what you’ve learned how to do this. So I love that we went deep on all these things. Just final question, is there anything else that we haven’t shared that you think might be helpful for listeners to hear or maybe just to double down on before we get to our very exciting lightning round?
Tomer Cohen: Whether you’re in an organization, you’re waiting for your leader to roll this out or you’re a leader trying to roll this out, I would not wait. The first thing I’ve done, which I thought in retrospect was very hopeful is I did announce this upfront we are going to this mode. We’re starting in pockets, we’re starting in pods, we’re building the tools, but this is the mountain we’re going to go after, and in many ways, we’re going to make it great. I also announced that this is not just an end state, it’s a kind of continuous progress. There’s no state we’re going to get to as much as continuously just trying to be better. And in many ways, to compete, you just want to be better than others in how you build because the version of building will completely just transform itself every few years or so.
So do not wait. Really focus on the progress you’re making, over communicate with your team, not just the vision, but also the progress you’re making, almost like holding yourself responsible. If you’re a leader, give yourself KPIs you share with your own teams or OKRs. And if you’re inside of the organization, and I would say whether or not or not your CPO or your CEO is announcing this type of program, go do it or join an organization that does it so you can be at the cutting edge of how you build in the future.
Lenny Rachitsky: Tomer, with that, we’ve reached our very exciting lightning round. I’ve got five questions for you. Are you ready?
Tomer Cohen: I’m ready.
Lenny Rachitsky: First question, what are two or three books you find yourself recommending most to other people?
Tomer Cohen: I love to give trios of books that I really like. So my current trio is, they’re very diverse in topics, so apologies if it’s not falling all into tech. But the first one is called Why Nations Fail. It’s a book I read a decade ago even more and the authors of it just won the Nobel Prize last year. And it basically talks about why does some nations succeed and some fail? And it’s not the usual explanations we go for, which is, oh, it’s culture, it’s natural resources, it’s the kind of religion. A lot of those tends to be the kind of immediate excuses people have. It kind of falls into two camps. Are there extractive or inclusive institutions? Can people participate broadly and opportunities shared or there are institutions that basically are supposed to be attracting from many and give to some.
So it’s just an incredible way to just think about how you build a nation. And for us at LinkedIn, we think a lot about the idea of opportunities, so how you build a product as well. And it’s just a good way to move away from easy explanations into what really makes a country really successful as well. Second book, it’s called Outlive. It’s really about the idea, it’s kind of like the author, Peter Attia talks about the idea of medicine 3.0, which is really the notion of building personalized medicine, which I think in the world of AI will become incredible in the future. But it’s all those, let’s call those categories that you should think about for your life so you can just optimize your health as much as possible and goes for everything through fitness to diet to the biggest health factors you should think about. But it’s a great long book. Then lastly-
Lenny Rachitsky: The one in my bookshelf behind me.
Tomer Cohen: There you go.
Lenny Rachitsky: It’s up top. You can’t actually see it, I think.
Tomer Cohen: And then lastly, it’s a book that also came out many years ago, but it’s called The Beginning of Infinity, which I really like, by Deutsche. It wasn’t an easy read for me, but I love the idea. In fact, especially in products, I love the idea of cause and effect, like really finding great explanations for why things happen and then building on top of that your next iterations. And this book really pushes on the idea of explanations that only once we have a clear understanding of what things happens, then we can have breakthroughs on top of that. But until we get to a point of clear scientific breakthroughs, we are not going to make significant progress. But when you do that, it’s really almost like infinite progress you can make on top of that.
Lenny Rachitsky: Naval’s always talking about that last book. I think I bought it and it was just hard reading this.
Tomer Cohen: It’s not an easy read, at least for me. It wasn’t an easy read, but it’s a very powerful read.
Lenny Rachitsky: Awesome. Is there a favorite recent movie or TV show you really enjoyed?
Tomer Cohen: Can I do a podcast?
Lenny Rachitsky: Absolutely.
Tomer Cohen: So there’s a podcast in, it’s in Hebrew, it’s called One Song, and it takes a song that generally is ideally popular and then goes really deep on the origin and the history of the song, and I love it. I love music and just dissects songs so well. It does a great job also in bringing to life the story behind it. So for me, it just goes back to you thought the song was about something, but then it goes really deep into the actors behind the song, and sometimes it’s the words chosen or it’s how the lyrics match the music itself, and I just really enjoy that one.
Lenny Rachitsky: There’s a podcast called Song Exploder, I believe, that is a similar concept that’s not in Hebrew, in English, that I’ll point people to if you love that one.
Tomer Cohen: That’s awesome.
Lenny Rachitsky: Is there a product you’ve recently discovered that you really love? Could be an app, could be some clothing, could be a kitchen gadget, type gadget.
Tomer Cohen: Can it can be a product I want to have, which I think is actually really easy to do?
Lenny Rachitsky: I love that. This is a product thinking 101 and just the vision of what you want to see.
Tomer Cohen: So in my car right now, there’s Alexa built-in, which is great because the kids can ask for songs all day long and it’s a whole show inside of the car. But one of my favorite things to do when this has been doing it for well over two years is I go in and I go into voice mode.
Lenny Rachitsky: ChatGPT.
Tomer Cohen: Yeah, ChatGPT, and then just have a conversation, and that’s just friction. I would love to have on my steering wheel a button that invokes my AI friend that can sit next to me in the passenger seat, and I just think that would be such a… I actually think it would [inaudible 01:01:36] rides for people. Just that movement, that’s just like elimination of friction will transform the experience for me.
Lenny Rachitsky: On that note, I recently discovered Teslas actually do this now. If you hold the right wheel, Grok appears and you could talk to Grok. So it’s here. The AI has arrived. Yeah. I just did it by accident and then it’s, “Okay, cool.”
Tomer Cohen: Great. So for me, if anybody from Rivian is listening, please bring this in the car.
Lenny Rachitsky: Rivian’s falling behind. Yeah. And you have to use Grok. It’d be cool if you could switch to different AIs because it has a personality. Just give me information. I don’t need you to laugh and give me jokes.
Tomer Cohen: Did you need to spend some time with it before or did it have any memory from… Did you bring any memory into it?
Lenny Rachitsky: There’s a logged out version and then you could just log in and it connects to your account. Yeah, it’s extremely cool. No one’s talking about it. It’s crazy because I don’t know if they launched it fully, but it just appeared.
Tomer Cohen: Do you talk in the car a lot to it?
Lenny Rachitsky: I don’t use it that much, to be honest, but I should. My wife just doesn’t love Grok. I think the brand of Grok is a specific brand. And so she’s like, “Don’t talk to Grok in here with me.”
Tomer Cohen: I love voice mode, so I use it all the time.
Lenny Rachitsky: Yeah, I love voice mode too. It just interrupts too often. That’s the issue there, right? It’s just it stops.
Tomer Cohen: By the way, you can set it up. You can basically say like, “Hey, just let me finish.”
Lenny Rachitsky: I now know that. I’m learning so much. Okay. Two more questions. Do you have a life motto that you often find useful in work or in life?
Tomer Cohen: I think last time I talked about it, I most associated here with, I might be wrong, but I’m not confused, although I don’t say it as much anymore. But I think the one I love, growth mindset is a second religions for us at home. And one thing I love about, there’s a phrase there that is becoming is better than being, which I think ties into the FSB mode a little bit, which is you’re always in progress mode, iteration mode. It’s not about reaching a state. It’s about the journey, the process. That’s what you should fall in love with. It’s about continuously growing and evolving without the negativity of it or there’s no sense of FOMO there. It’s just this continuous thing. If I look back a year from now and I look back, how much did I grow? How much do I know? What skills to do that again? Where are I becoming better? Do I feel like Tomer version 2026 versus 2025? What’s the delta there? And I kind of love that as a way of thinking.
Lenny Rachitsky: A great segue to our final question. By the time this episode comes out, it won’t be a secret that you’re leaving LinkedIn after 14 years. Legendary run. You joined way before the acquisition, you helped them integrate. Just like the way LinkedIn was perceived 14 years ago is so radically different from the way it is today. It’s actually really fun and interesting to be there versus how people for a long time felt about LinkedIn. So I guess the question just how you feeling and what’s next? I imagine you’re going to get a lot of calls from a lot of people, but what are you planning?
Tomer Cohen: Yeah, so I feel proud. It’s been an incredible ride at LinkedIn. The way I’ve got to know about LinkedIn deeply the very first time was when I moved to the Valley and I went to a lecture at Stanford about social networks in 2008 and Reid was there and he talked about the power of being a professional communities online, and I was very nerdy about it and thought it was incredible vision, had no plans to join and actually started my own company after. But as luck would have it, found myself joining a few years after and just thought the mission was incredible. So in many ways it aligned with my purpose and just was an incredible ride to be here.
And I also feel very grateful. I shared this with the company recently. I was starting to take learnings from my experiences here. A lot of it was from tough situations. We had a lot of tough situations at LinkedIn and hard calls and late nights, but you learn so much from those and I’m just incredibly grateful. And I’m excited. I’m excited. I have a bias for change. I have a bias for kind of positioning myself in a place where I can learn the most and learn a lot. And it’s an incredible time to build, so I’m just excited to be thinking of new problem sets and new areas where I can go deep on and invest the next decade in.
Lenny Rachitsky: I think it’s going to take a long time for you to not feel like you’re working on LinkedIn and to forget about all the things that you have been worrying about for so many years.
Tomer Cohen: After you build something for such a long time, and I think you and I talked about it at one point, that I think one of the best traits for a builder is to become very passionate with what they’re building. Really care. Not about the job. It’s really care about the product. When you feel the pain when somebody complains and you kind of have this continuous discontent, and it’s like for me, it’s the notion of raising a baby. So yeah, it’s hard. It will be hard. I will always think of LinkedIn as one of the babies I helped grow.
Lenny Rachitsky: Well, I’m excited to have you back someday when you figure out what you want to do next and or start whatever you’re doing. I love that this was an excuse to get to know you. Tomer, thank you so much for being here.
Tomer Cohen: It was great to be here. Thanks, 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.
Reformatted by reformat_english_direct.py
在AI浪潮下,产品管理正经历一场深刻的范式转移。LinkedIn首席产品官Tomer Cohen指出,到2030年完成工作所需的技能将发生70%的改变。面对变化远快于响应的现状,传统产品开发中不断叠加的流程与组织复杂性已成为创新的桎梏。为此,LinkedIn正试行“全栈构建者模型”,通过人机协作打破职能边界,让构建者直接将想法推向市场。本文并未泛泛而谈AI的替代效应,而是深入剖析了如何利用AI重拾产品构建的工匠精神。Cohen指出,仅提供工具无法驱动变革,配套的激励与操作范例才是关键。对于思考如何重塑团队效能的管理者而言,本文提供了一份极具前瞻性的实战蓝图。
为什么AI正在颠覆传统产品管理 | Tomer Cohen (LinkedIn CPO)
Tomer Cohen: 当我们审视完成工作所需的技能时,到2030年,它将发生70%的改变。因此,无论你是否打算换工作,你的工作本身都在发生变化。为了保持竞争力,你实际上必须回归一些第一性原理(first principles),回到绘图板前,重新构想构建产品究竟意味着什么。
Lenny Rachitsky: 你正在LinkedIn尝试一种截然不同的产品构建方式,它充分拥抱了AI所释放的潜力。
Tomer Cohen: 我们称之为全栈构建者模型(full stack builder model)。其目标本身是赋能优秀的构建者,让他们能够将自己的想法推向市场,无论他们扮演什么角色、处于哪一层技术栈,或者隶属于哪个团队。这真的是人与机器之间非常流畅的互动。
Lenny Rachitsky: 这感觉像是可能会成为许多公司未来运营方式以及产品最终构建方式的一种模型。
Tomer Cohen: 这里的变革管理将成为关键部分,但仅仅为他们提供工具是不够的。你必须构建激励机制、动力,以及如何操作的具体范例。我看到很多公司推出他们的智能体(agents),然后就期望公司能直接采用。这种方式是行不通的。
Lenny Rachitsky: 一直以来都有这样一个问题:AI是只会让那些不够出色的人变得出色,还是会让那些已经非常出色的人变得更出色?
Tomer Cohen: 顶尖人才有一种不断试图精进其技艺的倾向。我要为构建者强调的关键特质是……
Lenny Rachitsky: 今天,我的嘉宾是 Tomer Cohen,LinkedIn 长期的首席产品官,他正在试行一种新的构建方式,我认为这将成为未来公司运营的模型。它被称为全栈构建者项目(Full Stack Builder Program),本质上其理念是让任何人,无论其职能是什么,都能将产品从想法推向发布。他们废除了 APM(助理产品经理)项目,并用助理全栈构建者项目取而代之。他们引入了一条新的职业路径,头衔是全栈构建者,任何职能的任何人都可以成为这样的人。正如你将在对话中听到的,他们构建了一系列内部工具、智能体和流程,基本上是建立一个人机协作的产品团队,能够非常快速地推进、迅速适应变化,并以少得多的资源做得多得多。如果你正在寻找关于如何重新思考团队运营方式,并倾向于AI为团队和公司释放潜力的灵感,这期节目就是为你准备的。
重塑产品构建的原因
Lenny Rachitsky: Tomer,非常感谢你的到来,欢迎来到播客。
Tomer Cohen: 谢谢。很高兴能回来。
Lenny Rachitsky: 很高兴你能来。我非常兴奋能进行这次对话,因为你正在 LinkedIn 尝试一种截然不同的产品构建方式,它充分拥抱了 AI 所释放的潜力,倾向于利用现在成为可能的事物。对我来说,这感觉像是可能会成为许多公司未来运营方式以及产品最终构建方式的一种模型。有许多产品领导者都在谈论 AI,谈论它能做什么。但感觉你是真的在以一种非常、非常激进的方式践行这一点,所以我非常期待能向你学习,听听关于这些的见解,让听众了解你的所见所闻以及你学到了什么。让我先从这个问题开始:你为什么认为这是必要的?为什么你要重新思考这些关于产品长期以来如何构建的事情?换句话说,为什么人们需要关注我们接下来要谈论的内容?
Tomer Cohen: 这确实要从最基本的东西说起。对我来说,技术始终是关于赋能的。它不在于它为我们做了什么,而在于它使我们能够做什么。现在,在我看来,我们拥有这个绝佳的机会,使其成为一种精英体制(meritocracy),我认为这是一个机会,但此刻它也是一种必然。我想把它放在这样一个背景下来看:我们正在进入这样一个阶段,变化的时间常数远远大于响应的时间常数。这基本上意味着变化发生得比我们能够响应的速度还要快。现在,LinkedIn 对工作世界有着独特的视角。因此,我们实际上有一些在我看来令人震撼的数据来让这变得直观。当我们审视完成你的工作所需的技能时,到2030年,也就是字面意义上从现在起的四年时间,听起来似乎很久,但也就是四年后,它将发生70%的改变。
Tomer Cohen: 因此,无论你是否想要换工作,你的工作都在发生变化。唯一的问题是,你是否能保住它?然后从组织层面来看,目前增长最快的工作,市场上需求最大的工作,比去年增长最快的工作增长了70%以上。因此,作为一个组织为了蓬勃发展所需的条件,出现了一种新的迭代。
然后当你将其应用到产品构建上时,你会意识到,为了保持竞争力,你实际上必须回到一些第一性原理(first principles),回到绘图板前,重新构想构建到底意味着什么。而我非常喜欢这一点,因为当你思考构建者的角色时,构建者是公司的核心,目标其实非常简单。构建者提取一段DNA,将其赋予生命。这真的是整个过程,对吧?
我们都建立了这些,姑且称之为最佳实践。你非常深入地研究问题,你制定规范,你设计它,你编码,你发布,你迭代。基本上就是这样。但在许多规模化的公司中,包括LinkedIn和许多其他公司,随着时间的推移,这个过程变得非常快地复杂起来。
那么发生了什么?我们采取了每一个步骤,并将其扩展成了许多子步骤。研究问题变成了对我们来说查看10到15个信息源,显然还要与客户交谈,进行数据池分析,查看多个来源的反馈工单,社交媒体,与客户的互动。在我们觉得把问题研究得非常、非常透彻之前,我们大概要查阅10到15个信息源。
想想产品的评审。有设计评审、隐私评审、安全评审。我可以一直列举下去。每一个子步骤实际上都有其存在的合理理由。但当你把这一切加在一起时,你会想,“我的天。这就是为什么构建一个小功能,需要多个团队、多个代码库、多个冲刺才能把它发布出去”,更不用说迭代了,而这实际上才是你寻求成功的地方。
你永远无法在发布本身看到成功。所以实际上工作本身并不复杂,而是我们让过程变得非常复杂。当我深入挖掘时,我发现这还没有结束,因为必须有人来做所有这些子步骤,所以发生的事情是,你实际上也从流程复杂性走向了组织复杂性。
然后你实际上导致了微观专业化。所有这些子集都由特定的人在做。因此,从一个构建者,我们有了多个职能。显然我们有工程、产品和设计,你可以开始质疑这些界限。至少我在内部是这样的。
然后从那里,我们有了很多子专业。这发生在每一个职能中,但想象一下设计。我们有交互设计、动画设计、内容设计、研究。这方面有太多的维度。所以它们都是合理的,但它们都有对应的人员,而整个过程基本上意味着大量的……这基本上就是臃肿。这就是复杂性。
然后在不经意间,你最终得到了这个极其复杂的……我们实际上有这样一个图表,基本上展示了流程复杂性和组织复杂性结合在一起。通常人们会感到震惊,因为他们只在一个非常具体的事情上工作,但当你退后一步看时,你会面对这种压倒性的体验,让你不禁开始思考。而现在我们有了这个真正的机会去收拢堆栈,回归工匠精神,重新思考产品开发生命周期,这就是全栈构建者模型(full stack builder model)诞生的地方。
Lenny Rachitsky: 哇。好的。这里的信息量太大了。在你讲述时我们会展示视觉图表,以帮助人们看到你在这里解释的内容。而所有这些都非常理性。如果你有15个信息源,为什么不从中提取呢?为什么要错过那些东西?
你在这里描述的是,随着你获得更多的权力和更加专业化……这在理性上都说得通,但当你开始退后一步看这些时,就像,我的天,上线一个功能居然需要六个月。我想问问你分享的那个数据。我认为这是一个极其有力的数据,而且你在这里有非常独特的数据来告诉你这类事情。所以你说,未来人们需要的技能中大约有70%将发生改变。
Tomer Cohen: 为了完成他们当前的工作。
Lenny Rachitsky: 为了完成他们当前的工作。这是基于什么得出的?仅仅是基于历史数据,还是你们怎么发现这一点的?
Tomer Cohen: 是的。公平地说,变化一直存在,对吧?所以从来就不是仅仅保留你今天拥有的技能,但我们从未见过你当前角色中如此剧烈的变化。因此,无论你现在是一名营销人员、销售人员、招聘人员,还是工程师。工程领域是目前在智能体(agents)方面投入大量资金的领域。这些工作将发生剧烈变化。
我记得我说过我的角色,我作为工程师的生活,即使在10年后也发生了实质性的变化,而我们现在看到的变化,仅仅想想四年后,要真正做得非常、非常好的工程工作将会截然不同,或者构建软件,构建某种产物。但这几乎适用于每一个职能。它不是平等的。像护士这样的工作受到的影响会较小,但有些工作会看到90%、95%的影响。
Lenny Rachitsky: 还有一个数据,我觉得你在这里没有提到,是我在你第一次谈到这个项目时的帖子里看到的,那就是今天70%增长最快的工作甚至不在一年前的工作列表中。
Tomer Cohen: 是的。没错,所以列表上增长最快的工作在一年前还不存在,而且其中许多在一二十年前甚至都不存在。实际上跨领域来看有一些相当惊人的数据。
全栈构建者项目
Lenny Rachitsky: 好的。那我们来谈谈你构建的这个项目。告诉我们它的名字,然后告诉我们它目前的要点,以及你对它未来愿景的期望。
Tomer Cohen: 是的。所以我们称之为全栈构建者模型。目标,总是从目标开始。目标本身是赋能优秀的构建者将他们的想法推向市场,无论他们的角色、技术栈以及具体在哪个团队。最终的想法是让那个构建者能够端到端地开发体验,将传统上不同领域的技能和专业知识结合起来,把它们融合在一起。这不是一系列步骤的顺序。它实际上是人与机器之间流畅的互动。这就是我看待它的方式。
然后当你回顾那个产品开发生命周期,从想法、洞察一直到上线,我为构建者强调的关键特质是,我希望他们把时间花在哪里,也是我认为优秀的构建者应该大放异彩的地方。所以是愿景的概念。想出一个关于未来的引人注目的构想。同理心,超级关键,对吧?对未满足的需求有深刻的理解。
沟通能力很关键。我们现在在几乎每个角色的职位描述中都看到了很多这一点,但你围绕一个想法去与他人对齐并凝聚他人的能力。创造力,对我来说就是想出超越显而易见的可能性。例如,我不认为AI目前在创造力方面很出色。我认为它在很多方面,带回了一些你可能不知道的东西,但它不是那种更高层级的创造力,我认为人类在这方面仍然要好得多。
Tomer Cohen: 然后最终我认为构建者最重要的特质是判断力。有些人称之为做测试(test making),但这其实是在复杂模糊的情境中做出高质量的决策。其他的一切,我都在非常努力地自动化。真的,非常努力。然后当你思考结果时,这不仅仅是为了获得更多的射门机会,我认为人们会说,“哦,迭代速度会非常快。”是的,但你实际上对大规模组织所做的是,它们变得非常敏捷,更有适应性,也更有韧性。它们能够驾驭未来。它们实际上能够将变化的步伐与响应的步伐相匹配。我脑海中有一个类比,有点像海豹突击队。你来到训练营,他们都在学习,他们在多个领域接受交叉训练。他们专长于任务,他们以小组(pods)形式运作,非常敏捷,你可以很快地将他们集结起来。我认为这将是未来会获胜的组织形态。
全栈构建者的愿景
Lenny Rachitsky: 好的。所以简单的想法,如果你只用一句话来概括,这里的意思是有一个构建者基本上独自走过整个产品开发过程。他们有一个想法,他们做调研,他们看数据,他们做原型设计并发布。这大概是这个方向的发展愿景吗?
Tomer Cohen: 是的,但不一定非要独自完成。这不是说……我仍然相信团队。
Lenny Rachitsky: 明白了。所以是更小的团队。
Tomer Cohen: 就是更小的团队。更小的团队,并且更加专注于问题和任务,而不是……实际上,作为例子,我们做的事情之一是开始引入小组(pods)的概念。我们不再是大团队。我们组建一个团队,理想情况下是全栈构建者聚集在一起,这不再是我能不能让一个工程师、一个设计师和一个产品经理合作并试图组合这个三人组,而是寻找能够跨界灵活应变的人,然后他们花一个季度左右的时间解决某个问题,接着我们重新组装成不同的组。这是我们现在正在做的一个实践例子,实际上在速度方面,以及在团队的专注度和敏捷性方面,都看到了巨大的成功。
Lenny Rachitsky: 感觉这里的目标,你试图调整的,以及当团队臃肿时断裂的,是速度、适应性和灵活性,因为回到你最初的观点,现在变化发生得快得多,以至于以传统方式构建的公司根本无法竞争。
Tomer Cohen: 是的。并不是你必须打破这个模型。我认为这个模型已经破裂了。只是这种变化的步伐帮助我们意识到了这一点。
Lenny Rachitsky: 好的。那么回到这些构建者仍然要做的事情与你想自动化的事情。你分享的清单是他们负责愿景、同理心、沟通能力、创造力和判断力。
自动化的三个支柱
Tomer Cohen: 是的。是的。我会把很多重点放在后者上。我认为如果你在一天结束时问我,哪种特质最重要?我会说是判断力,那种做测试的能力。
Lenny Rachitsky: 那么在你要自动化的方面,你在哪些领域看到了真正自动化的巨大成功,你认为这会走向何方?
Tomer Cohen: 是的。所以我想把它拆解开来,我认为这是……如果你现在是一家初创公司,在很多方面你可以从这种方式开始。没有遗留代码,没有你运行的遗留结构。事实上,我交谈过的许多原生构建AI的初创公司,他们就是按照全栈构建者在工作。这就是他们起步的方式。如果你处于像我们这样规模的公司以及市场上的许多其他公司,你会觉得,这几乎是一种你必须去做的全新生产函数和思维方式。我们真正在努力的有三个组成部分。一个是平台。第二个是工具和智能体。最后是文化。
平台这一块,这是你必须在此之前进行的那种投入水平,在它真正开始之前,你开始看到所有收益的累积。但以我们为例,平台是重新架构我们所有的核心平台,以便AI能够对其进行推理。所以我们正在构建这种带有我们实际构建的服务器端的可组合UI组件。我们基本上是在为AI做好准备以便将其引入。所以你不能只是带一个第三方工具进来就能让它在LinkedIn的技术栈上工作。事实上,这是我们最大的教训之一。它从来都不起作用。从来都不起作用。你必须引入它并对其进行大量定制,几乎以alpha模式与那些公司合作,使其在内部运行。
Lenny Rachitsky: 所以这本质上是为了让AI更高效地工作而重新架构你的代码库。可以这样理解吗?
Tomer Cohen: 是的。而且在很多方面,是与那些公司合作,调整他们技术栈中的一些东西,以便也能与我们的技术栈配合。
Lenny Rachitsky: 当你说那些公司时,意思是像Cursors这样的开发智能体以及[听不清]之类的吗?
Tomer Cohen: 是的。或者是设计方面的Figma。或者你可以把设计系统(design systems)看作是另一个例子。但你必须要有这种来回沟通,因为他们不是……在很多方面,我们还没有看到任何人能够立即开箱即用地在我们基于代码的设计系统和我们拥有的独特上下文中工作。
Lenny Rachitsky: 简短地顺着这个线索问一下,所以有Figma。这很有趣。所以基本上Figma导出和维护你的设计系统的方式,必须改变才能更好地与AI配合,这是我听到的。
Tomer Cohen: 他们首先需要知道如何与我们的设计系统配合,这在很多方面是那些公司正在努力的事情。编码也是一样。你不可能只是把它带进来,它就能很好地对你的代码库进行推理。我们试过了。我们正在构建那个基本上允许它这样做的层,无论是Copilot、Cursor还是Windsurf等等。
Lenny Rachitsky: 明白了。好的。哦对,Copilot。微软。我懂了。我懂了。好的。好的。所以那是平台。所以这是你们必须做出的一项投资,以使AI在构建和做所有这些事情上变得有效。
工具与智能体
Tomer Cohen: 然后你有工具。工具是你真正构建智能体的地方。我提到过我想自动化我们谈到的那五个特质之外的所有事情,然后我们正在为此构建工具。而且对于那一点,实际上非常相似,我不能只是从外面带一个工具进来就用。所以我给你举个例子。我们最重要的事情之一是构建一个信任智能体(trust agent)。信任对我们在LinkedIn来说非常重要。有很多独特的维度,信任在LinkedIn扮演的角色在其他任何地方都不存在。所以我们需要将所有这些诀窍、上下文和信息库带入那个智能体中。所以我们最终在LinkedIn构建了我们自己的信任智能体。
Lenny Rachitsky: 那么这个信任智能体在做什么?告诉你什么时候你可能正在暴露你不应该暴露的信息吗?
Tomer Cohen: 所以当你构建一个规格说明,构建一个想法时,你走过信任智能体,它基本上会告诉你你的漏洞是什么,你潜在引入的或因此将被引入的伤害维度是什么。我是让我们的信任负责人来构建它的。所以每个领域的领域负责人都在构建他们自己的智能体。举个例子,我们为求职者提供的一个功能叫“Open to Work”(公开求职)。如果你在找工作,你可以放一个公开求职的标志。
Lenny Rachitsky: 是的,头像圆圈上的一个绿色加载小东西。
增长智能体与研究智能体
Tomer Cohen: 没错。实际上这是一个很好的信号。我从中看到了一些很棒的成功案例。人们在互相帮助,社区在互助中真正繁荣起来。但与此同时,因为它代表公开求职,这也为恶意行为者引入了一个信任向量。找工作的人可能比其他人更容易受到诈骗。因此能够提前思考如何预防所有这些情况。所以我们把几年前的那个规格说明交给了信任智能体。它不仅能够找出我们最初发起的所有内容,还找出了我们直到后来才发现的漏洞。所以这是一个真正运作得非常好的例子。
这是一个例子。另一个例子是增长智能体。再说一次,LinkedIn 有一个非常独特的……实际上,我们有一个令人难以置信的增长团队和增长流程。我们基本上把所有独特的循环、漏斗、过去的测试,所有东西都输入到了这个增长智能体中,现在你基本上可以把你的想法放进去。它不仅能让你的执行变得更好,它实际上会评判你的想法有多好。这是你无法直接买现成的东西,它是 LinkedIn 非常独特的。所以我们不得不在其中进行大量投资。现在正在使用它的一个团队,几乎……我一开始没有想到,但我们的 UXR 团队,也就是 UER 团队,用户研究团队通常在使用那个增长智能体,去了解在所有基本上为会员呈现的内容中,哪一个具有最大的增长机会和最大的影响力?当我们思考那个想法时,这并不在计划之中,但团队基本上都在把那些想法输入到这个智能体中。
另一个例子是我们的研究智能体。研究智能体基本上是在我们会员的角色画像上训练的。你可以想象一个小企业主、一个求职者等等。它使用的不仅仅是世界知识,它使用的是我们过去所做的所有研究,所有进来的支持工单。所以它非常擅长理解 LinkedIn 上的那种角色画像。所以我们有一个例子,一个团队提出了一个规格说明。他们当时还不知道我们有研究智能体。我替一个小企业主向研究智能体提问,想让它思考我们的营销规格说明,它评判得非常出色。实际上,在很多方面它改变了团队的方向,让我们专注于我们可以关注的其他集成工具,但是想要在公司内部看到所有这些知识库的全貌是非常困难的。
这是另一个例子。我们有一个分析师智能体,训练了如何基本上查询整个 LinkedIn 图谱,那个图谱非常庞大。你可以使用分析师智能体,而不是依赖你的 SQL 查询或数据科学团队。所有这些我想说,我会称它们仍然处于 MVP+ 阶段。我们在接下来几个月的目标基本上是在外部推广它们。所谓外部,我的意思是在 LinkedIn 内部。
Lenny Rachitsky: 不是作为新的产品线。 Tomer Cohen: 没错。 Lenny Rachitsky: 好的。有很多问题。一个就是你们是如何构建这个的?有在使用某个平台吗?在 LinkedIn 构建一个智能体需要什么?都是内部工具还是有使用第三方?
智能体的构建平台与工具
Tomer Cohen: 问得好。所以我认为我们一直在试验很多工具。我想说,对于很多那种知识库智能体,我们使用了从 Copilot Enterprise 到 ChatGPT Enterprise 的各种工具。然而到目前为止,最重要的部分基本上是我们自己的定制。那是我们看到最大收益的地方。甚至是在这些工具之间构建协调器,因为你希望智能体开始相互跟随,信任智能体应该与增长智能体一起工作并来回互动,而不是更多地进行顺序操作。所以我们在内部做了很多工作来实现这一点。这就是为什么我认为它确实需要那种程度的投资。
然后在某些情况下,让我们谈谈我们正在使用的设计智能体。我们正在与多家公司合作,试图了解哪种产品最适合我们。有趣的是,这是另一个经验教训,不同的团队被不同的产品所吸引。所以这是我们必须要解决并思考如何做得非常好的事情,因为最终我们试图尽可能地简化流程,但这对我们来说是一个很大的学习,关于我们使用哪些工具以及我们基本上如何将它们整合进去。
Lenny Rachitsky: 明白了。所以你可能有一个出色的 Figma 智能体,但有些团队想使用不同的设计工具。 Tomer Cohen: 是的。所以我们有点试验了 Figma、Subframe 和 Magic Patterns 等等,我们看到人们根据他们的职能、他们的可见度水平、他们以前对工具的了解程度,被吸引到不同的工具。最终,我不想在公司里有八个设计智能体,所以我们必须至少收敛到几个。我认为在很多领域都是类似的,因为那些智能体的吸引力在于,很多智能体试图解决相似的最终目标,但它们的做法非常不同。你会看到最终,我不认为会出现赢者通吃的情况,因为客户或用户的起点将在很大程度上决定它们对于那个用例来说有多简单。
单一职责与编排层
Lenny Rachitsky: 非常有趣。这里另一个有趣的收获是,你们正在设计非常特定的、只做一件工作的智能体。这是一个非常深思熟虑的决定吗?你们有没有尝试过一个在所有这些事情上都超级智能的智能体?
Tomer Cohen: 最终,它们会有一个协调器。我们将在跨领域真正做协调,但我们确实希望能够非常好地对这些智能体的表现进行评级和打分。而且我认为存在一种专业水平。现在,我们正在以一种能够掩盖很多这些细节的方式来构建它。你可能不知道有一个信任智能体。你可能有一个我们内部称之为 product jammer agent 的东西,它基本上做你的 product jam,这是我们内部做的一个流程。你可能只是使用 product jam 引擎,然后那个 product jammer agent 会与所有其他智能体一起工作。但现在我们从这些构建块开始,直到我们在跨领域构建出编排层。
投资重心与编码加速
Lenny Rachitsky: 从你分享的内容中另一个有趣的收获是,大量的工作投入到了产品开发流程的开端,就像帮助你制定正确的要求,澄清信任问题,然后这里是 product jam,这里是我们做过的研究。我想这是因为编码已经通过所有这些 IEE 工具得到了加速。谈谈为什么这可能是大多数投资投入的地方,以及到目前为止你在哪里看到了最大的影响。
Tomer Cohen: 嗯,我们 100% 的编码投资已经投入进去了,这开始于一段时间以前,并且那些都已经到位了。我们有我们的编码智能体。事实上,我们有点把它分成了两个部分。有从想法到设计的部分,然后我们称之为从代码到发布的部分。从代码到发布的部分受到了很多关注,我们在这方面取得了很大的进展。从编码智能体到当你构建失败时我们称之为维护智能体的东西,它会为你处理。事实上,我认为我们接近 50% 的所有这些构建是由维护智能体和一个 QA 智能体完成的。
Lenny Rachitsky: 哇。所以这是当构建中断时,而不是工程师跳去处理那些问题,而是由智能体来修复。 Tomer Cohen: 你仍然可以去喝完你的咖啡,然后再去重新进行构建。 Lenny Rachitsky: 酷毙了。
但在想法到设计这个领域,直到我们推出这个项目之前,我们并没有太多投资。这也是工作中很大的一部分。至少在进入编码阶段之前,这也是大量工作质量的来源。目的是赋能每一个人。所以如果你是一名工程师,你基本上可以在流程的前端使用所有这些工具,并能够成为一个全栈构建者。
Lenny Rachitsky: 把这些东西建立起来花了多长时间,让你能够真正组建第一个团队来构建这些最初的智能体和一些后端,重做代码库之类的东西?
Tomer Cohen: 我在去年年底在内部宣布了这件事,我们真正开始工作,但更多是在内部建立团队和流程。我想说的是,在这些东西真正训练好之后,大概四到五个月,我们有了那些智能体的第一个 MVP。但实际上这项工作本身大概是几个月的专注工作。其中很大一部分是把所有的数据语料收集起来,进行清理。这实际上也是一个很好的经验。仅仅让它访问你的云盘然后说“在这个知识库上进行推理”效果并不好。它在理解过去事物的重要性以及对事物进行权重分配方面做得非常差。你实际上需要具体思考在什么上下文、什么时候想要提供给它,以及你希望它专注在什么知识库上。因此,即使是清理我们称之为黄金示例(gold examples 或 golden examples)以供学习,也成为了最大的经验之一。仅仅在你的整个知识库上进行推理是行不通的。
Lenny Rachitsky: 是的,这很有道理。可能就像某个研究员对某件事有强烈的观点,而你并不同意,但它并不知道。它会觉得,哦,当然,这是数据,这是事实。
Tomer Cohen: 完全正确。而且它并不总是能理解与最初规格说明的联系是如何导向成功的。你实际上必须构建……这是一种非常有趣的方式。当你思考如何引入这些工具时,你不能只是把它们引进来。你必须知道你用什么来喂养它们。而你喂养它们的内容不仅仅是访问权限。我看到很多人只关注连接和集成,这让我想起了……这几乎就像是,这实际上是十多年前,当时我正在联合重组团队,联合重组 LinkedIn 的信息流,我们从零开始,我必须真正坐下来过滤什么是 LinkedIn 上好的专业帖子,什么不是的示例。这就像是为了整理出那个黄金样本示例花了数周时间,但这并不是……最重要的部分是喂入正确的数据,而不是所有的数据。所以这需要工作。我想说的是,对于许多正在思考这个阶段的公司来说,我如今就这个流程与许多 CPO 和 CO 进行过交流。你必须投入这些初始工作才能在之后获得收益。当我思考这一点时,我认为在 AI 方面有一个普遍的启示,即使你是第一次学习它等等,无论是 Cursor 还是设计,无论是 Figma 还是其他工具或 Lovable,你应该准备好投入那些小时数,然后你才会看到自己在速度和质量上有所提升,这些提升会出现,但你必须投入那个时间。
试点现状与规模
Lenny Rachitsky: 试点的现状如何?规模有多大?有多少团队在做?你们发布了什么样的东西?请让我们了解一下目前的情况。
Tomer Cohen: 好的。我不认为我们已经达到了一个非常高的抽样率,也就是占组织很高百分比的程度,但我们已经有很大一部分组织在使用它并提供大量反馈。我们看到了很多很好的例子。所以我认为收益的方式是实验数量乘以质量的函数。也就是这些实验有多好,除以真正将它们做出来所需的时间,比如从想法到发布。所以在节省时间方面,我们看到了,无论是 PM、设计师还是工程师,现在每周都能节省数小时的工作时间,无论是我们谈到的分析师智能体,还是非常快速地进行原型制作,或者 product jamming 体验都是其中很大的一部分。在质量方面,我们看到洞察讨论变得好得多。顺便说一下,质量和时间,有时它们是相辅相成的,因为质量高,你就不必在某件事上花费那么多的时间。所以我们看到这一点正在被应用。至于数量,我不认为我们已经达到了我看到组织中很高比例的人在做的程度,但这会在我们……我们还没有在内部 GA(正式发布)这个。这将在接下来的几个月内到来,一旦我们把所有的东西都准备好。但我们看到设计师和 PM 直接从 Jira 工单中挑选 bug,把它们推进去,这是我们以前没有见过的,而且每个人都渴望加入。所以实际上,现在最大的问题是每个人都想要访问权限。每个人都想要获得这些工具的访问权限,以便能够一起做这件事,而我们只是想确保它足够好,以确保整个组织都能做得很好。
试点推广模式
Lenny Rachitsky: 那么你们是如何推广它的?是有一部分人拥有这些智能体的访问权限,他们只是在拥有这些工具访问权限的情况下按照他们原来的方式工作?还是有一个专门的团队,规定这就是你现在的工作方式,就是这样,然后看看会发生什么?
Tomer Cohen: 这是一个很好的问题。所以基本上我们有一个构建团队。这是跨整个研发部门构建 FSB 路线的核心团队,FSB,也就是全栈构建者。然后有一些使用它的小组和小分队。所以基本上我们正在寻找特定的领域,基本上把它交给他们。条件是他们要提供反馈。作为回报,他们让工具变得更好,所以这不仅仅是访问权限。我们想要那些会使用它的人。所以你的早期采用者之一将是那些帮助真正把产品打磨得很好的人。所以我们现在正在以小分队模式来做这件事。
Lenny Rachitsky: 所以这就像是一个更大团队中的小分队,比如由设计师、PM、工程师组成的小组存在于……有例子吗?你有一个尝试这个的 LinkedIn 部门吗?
Tomer Cohen: 是的。所以如果回想我们的一些团队,不管是不是……实际上,我们刚刚推出了语义人员搜索和语义职位搜索。那个团队就在使用这些工具的一部分来实际帮助构建它。所以那个团队实际上,是 PM 们用这些工具构建他们自己的仪表板,而无需等待设计资源的介入。然后我们有一个设计团队,他们现在……这实际上是从经理推广开始的。在很多时候,我告诉这个团队的是:“不要等待正式的 GA(通用可用性)。开始做吧。开始投入吧。”我们看到那个团队的设计师开始提交 PR(拉取请求),这以前从未发生过。现在其他团队也想要这样做。所以它是从这种草根体验开始的。我想说那些地方一直非常正式。我想说开头是在顶层的。
产品高管团队,基本上我们从职能领导者、设计、PM、BD(业务发展)等等,转移到产品领域领导者,他们基本上横跨技术栈大显身手,并且他们也会对所有这些职能进行 360 度的全方位评估,看看他们是否真的能够进行全栈构建体验。然后我们也在初级层面启动了一个名为助理产品构建者项目(Associate Product Builder Program)的新项目,基本上我们过去有我们的 APM 项目,这个项目今年就要结束了。然后从一月份开始,我们将开始我们的 APB 项目,他们将加入 LinkedIn。我们将教他们如何在 LinkedIn 编写代码、做设计和做 PM。他们将经历一个相当严格的培训过程,然后他们将加入那些小分队,我们将逐渐把这个项目发展成为 LinkedIn 的一个重要组成部分。
初级项目与全方位体验
Lenny Rachitsky: 哇。所以这可能是 APM 项目的未来,就是这种类似于 APM 的全栈构建者项目。
Tomer Cohen: 在很多方面,我们构建了一些非常令人惊叹的……我对那个群体感到非常兴奋。我希望我能加入它。但是我们为他们构建了很棒的培训。而且在很多方面,我们将利用这个培训来思考如何在整个组织中进行推广。我们有点像是在透过这样一个视角来看待:你拥有出色的技术技能,但你还不是公司里的工程师;或者你有很好的设计品味,但你还没有在公司里做过规模化设计,我们将教你如何在 LinkedIn 做到这些,而这个培训我们也会大量用于向全公司推广。
Lenny Rachitsky: 好的。所以你们有这些项目,这些试点和这些小分队,而且你说你用来观察这是否值得推广的指标是实验速度乘以质量乘以时间。
Tomer Cohen: 除以时间。
Lenny Rachitsky: 除以时间。好的。明白了。我猜我知道现在还为时尚早,但你说你看到它目前每周为团队节省了几个小时,大概是这样吧?
Tomer Cohen: 是的。而且我认为反馈是最重要的部分。对吧?理解这个的方式就像你构建一个产品一样。所以我们在内部构建这个产品,你希望与某种会给你反馈的早期采用者进行实验,而反馈一直非常惊人。事实上,我们的顶尖人才是 LinkedIn 上使用这个最多的人。来自他们的反馈非常不可思议,因为他们也愿意花时间给出反馈。他们在这方面的反馈也非常惊人,比如他们产出的质量,他们花在这上面以获得回报的时间,他们渴望成为其中一部分并真正扩展它、让它变得更好的意愿。所以很多兴奋感来源于他们如何使用它以及我们在那里看到的素质。我想说在六个月左右,我们将能够看到组织中有更多的人使用它,你会开始看到那些顶线数字也会随之增长。
顶尖人才的采用
Lenny Rachitsky: 这是一个非常有趣的洞察,表现最优异的人找到了最大的成功,因为一直有这样一个问题,AI 到底是会让那些不怎么样的人变得更强,还是会让那些已经很厉害的人变得更厉害?听起来很可能是后者。
Tomer Cohen: 是的。而且在很多方面,这令人惊讶,也不令人惊讶。我们在……的时候我也看到过这一点。令人惊讶是因为你希望其他所有人都能参与其中并投入进来。我认为顶尖人才有一种不断试图在他们手艺上变得更好的倾向,以及处于构建方式最前沿的内在需求,我认为我们在这里也看到了这一点。这就是为什么我对团队说过这样一句话,如果我们构建了所有这些工具,他们会用吗?我知道现在答案是否定的。仅仅给他们工具去使用是不够的。你必须构建激励项目、动力、如何去做的示例。他们也需要看到其他人取得成功。当我们将 LinkedIn 从一家桌面公司转型为移动公司时,我也看到了这一点。这是一个非常相似的过程。这非常困难。这里的变革管理将是一个关键部分。我认为我看到很多公司推出他们的智能体,只是期望公司去采用。这不是这样运作的。有些人会采用。那往往是你处于前沿的 5% 的人才,他们只是想要新工具,并且他们有变革的偏好。但绝大多数人需要在他们如何做这方面进行变革管理的工作,这需要在对文化层面进行更多的思考,对我来说这绝对是最大和最重要的事情。
变革管理与文化塑造
Lenny Rachitsky: 是的。我想在这方面多花点时间。人们不在这里花时间是很有道理的,因为他们有太多事情要做。他们得交付东西。他们的一天已经很忙了。你现在必须挤出时间来学习这个新工具,而它暂时还不会有回报。所以我理解为什么人们会说,“好的好的,我会弄的。总有一天我会用的”,但他们并没有。这种关于文化的理念,当我最初看到你分享这个时,这是使其成功的第三个部分。所以首先是让代码库为人们和 AI 协作做好准备的平台。然后是工具,就像你谈到的智能体,然后是文化。关于真正帮助人们上手的经验,你还能分享更多吗?我听到的一件事是制造一点 FOMO(错失恐惧症),比如,好吧,只有少数人可以使用这个,你必须注册才能获得访问权限。在让人们上手方面,什么方法是有效的?
Tomer Cohen: 是的。我认为这就是我向人们强调的地方,把所有事情做完,平台、工具,是不够的。这是它运作的前提条件,但不足以让它运作起来,因为它真的需要你在文化层面进行大量投资,即如何让人们投入其中。这在最初可能会感觉缓慢,但我以前在我们从桌面到移动的思维转变中见过这种情况。一旦它加速,实际上会保持非常高的速度。第一,人们真的会被你如何为他们定义期望所激励。所以想想某人在某个角色中的期望是什么,无论……
Lenny Rachitsky: 所以比如改变绩效考核之类的事情。
Tomer Cohen: 确实如此。从如何招聘到校准和评估,一切都在其列。我想在早期就看到的一件事就是这种 AI 智能体(agents)能力和熟练度。正如我提到的,工具就在那里。问题在于,你会使用它们吗?因为一开始这些工具会足够好用,但还谈不上出色。这是每一个优秀的 MVP 工具的经典特征。它们足够好用,但并不完美。然后你想要建立那种智能体能力来让工具变得更好。我们处于这样一种理念中,即我们要一起为 LinkedIn 把它变得更好。第二点是在组织内部试点成功。这就是舱室模型,你要展示的不仅是这行得通,而且确实取得了成功。因此,我们甚至让我们的合作伙伴团队、BD 团队能够直接上手,而不是依赖等待工程师来帮助构建开发者门户和其中的连接器。
实际上,我们的一位合作伙伴负责人就是亲自去做了。他甚至没有委派给他的团队。他们的目标是表达这样的意思:“嘿,我能做到。你也可以。”这些例子真的非常非常有力量。我谈到了助理产品构建者项目,我们将非常专注于培训。我认为这会在整个组织内传递一个非常强烈的信息。人们会看到这些人才以及他们能做什么,我认为这会创造那股浪潮。但要在全员大会上庆祝胜利,突出个人并展示那些例子。我们最近看到的一个例子,人们真的是带着惊讶的眼光来看待它,但随后我认为这确实为他们打开了一种视角。我们用户研究团队有一个人。我们在增长团队有一个 PM 的空缺,那个职位开放了一段时间,然后她说:“我觉得我能胜任。”
她使用了所有这些工具。这是一个用户研究员变成增长 PM 的例子,这通常不是你看到的职业路径,但她对这个领域感到兴奋。她使用了所有那些工具,现在她是团队里的一名增长 PM。而且实际上,你最终可以开始更多地把她看作一个全栈构建者。但看到这些机会,然后突出这两个人,实际上那些正在这样做的人已经成为了一个很好的榜样。然后只是确保那些工具是易于获取的。人们可以提供反馈,你分享很多,这已经是其中令人难以置信的一部分。仅仅靠自上而下的指令说这是我们想要的工作方式是不够的。人们想要感觉到有成功的故事。他们觉得这值得他们花时间。感觉这是他们想加入的一场运动,然后最终他们能在自己的做法中看到成功。
Lenny Rachitsky: 我很喜欢这种向移动端转变的比较。我们都经历过那个时期,有很多关于公司要求你展示移动端模型的故事。那是我们唯一运作的方式。现在你要交付的所有东西都必须在移动端上,这很有趣,它与那种经历是多么相似。所以你刚才分享的几件事,算是总结了一些对你有效的方法。展示胜利,庆祝胜利,向人们展示其他人在用 AI 工具做什么,创建一个让人们加入并带有一点排他性的项目。这个绩效评估的部分非常有趣,因为它真的会改变人们的行为。这是我们获得晋升的方式。你实际上已经对 PM 做出这种改变了吗?我想应该是每个序列,我设想不仅仅是产品管理。你已经做出了这种改变,还是说这像是一个正在进行的工作?
Tomer Cohen: 这其中有两个方面。一旦我调动了我的团队,我的直接下属,我们对他们做了 360 度评估。所以他们的 360 度评估是,如果你来自 PM,你团队里的设计师会给你打分。因此那有它自己的作用,然后我们把这些分享给他们,那有其自身的一种激励作用。但后来我们广泛地将其推广。所以现在我们招聘时,我们会寻找这些特质。然后在即将到来的周期中,我们做一次半年度评估。那将成为绩效评估的一部分,并且我们向所有人宣布了这件事。为了什么,这是人们兴奋地想要展示的地方。他们很兴奋地知道他们将如何……这总是关于,“我想知道我是如何被评级或评估的。”所以能够展示这些例子一直是其中的重要部分。
我想说的另一件事是,这个项目及其正式化在整个组织中推广是需要时间的,而且我故意不试图快速向所有人推广,因为我认为那真的会很快稀释它的价值,因为这不在于……我根本不在乎你的头衔。我在乎的是你如何工作。所以称呼你为全栈构建者并不是我想要的。将你的心态转变为全栈心态才是我想要的。你在想你可以完成整件事。你在看那些工具,在看如何去做。
所以我说过的一件事是,如果你在等待一次正式的组织重组或声明来开始以不同的方式构建,那你等得太久了。听着,我最大的一点是,这里有一个给我的许可,让我不用等待直接上手。所以不管你有没有合适的工具,去构建工具,使用外部的工具,把它引进来,展示那些例子。在许多方面,在其他任何事情想到之前,证明你在心态上是一个全栈构建者。而那自然就会发生,这也是我们看到一些最优秀的人才直接投入大量精力的地方。
鼓励自主探索与工具尝试
Lenny Rachitsky: 我喜欢这个。我本来实际上想提到那句话的。某个和你一起工作的人,你分享过,他告诉我的正好就是你刚才分享的那句话,所以我很高兴你提出来了,就是如果你在等待重组,你并没有用正确的方式思考它。你如何鼓励人们真正自己去把玩这些工具?你是像这样说的吗,“去花几天时间玩玩 AI?”只是去试试吗?或者有没有什么正式的做法,让你看到能让人们在不加入这个项目的情况下更多地自己去尝试?
Tomer Cohen: 我们制作的很多工具,我们一直在定期分享。我的几次全员大会都完全是关于如何使用那些工具的。但与此同时,我们有点像是在邀请,你有没有找到一个对你来说非常有效的新工具?分享它,展示它。再说一次,可以是 Slack,可以是 Messages,Teams 等等,看你是怎么做的。但这个想法真的是开始获得对事物如何运作的投入。实际上,我认为总的来说,你现在可能会对工具感到不知所措,对各种秘诀和如何做事情感到不知所措,比如什么是你的提示词,什么是我的提示词。但真正的关键是找到一些非常有效的东西,可以围绕它聚集并真正投入,这些就是那些领域。但我认为我们已经发出了这种邀请,去探索,去带来你认为很棒的东西。并且在很多方面,带上其他人一起踏上这段旅程。这是让影响力比少数几个在这方面做得很好的人大得多的一个好方法。
遇到的负面意外
Lenny Rachitsky: 这其中有没有出现什么负面的意外,比如 PRD 让人感觉就像是由 AI 驱动的,人们出乎意料地变慢了?有没有什么事情让你感到惊讶,比如,“好吧,这其实并不好”?
Tomer Cohen: 是的,我们提到了其中的几个。我本来希望一些工具开箱即用就能非常好。但这从来没有发生过,因为我们必须投入相当多。
Lenny Rachitsky: 从来没有过。
Tomer Cohen: 从来没有过。我们必须投入相当多。再说一遍,部分原因在于我们确实有大量的遗留信息、代码库、知识和设计系统等等。因此,与我们合作的许多公司也将其视为自身发展的绝佳增长机遇来进行投资,但我确实认为这也是一个巨大的投资领域。我们谈到过不能仅仅提供对所有上下文的访问权限,这是我们最初的尝试,当时我们觉得,“哦,这里可以访问所有的云端硬盘和所有信息”,结果一败涂地,并且产生了疯狂的幻觉。人们开始倾向于使用不同的工具,比如我们的目标是让工具收敛统一,但这非常困难。
然后我认为在质量方面,我们确实看到了更好的质量,但我认为这是因为,我们仍处于早期采用者阶段,他们在如何操作方面进行了几次迭代。但我想说,工具的采用是很困难的。然后我认为对一些人来说,我觉得有必要说明这一点,有些人不想成为全栈构建者,这完全没问题。有些人将自己视为专业化人才,我认为专业化有其存在的位置和作用。所以我不想向全组织传达的信息是我期望每个人都成为全栈构建者。我并不这么期望。我认为有赋能全栈构建者的系统构建者,然后你也有专业化的人才。但我认为我们不需要像过去那样多的专业化人才。
全栈构建者的职业路径
Lenny Rachitsky: 我直到刚才才意识到这一点。那么现在这是他们的头衔了吗,不再叫产品经理或工程师,而是全栈构建者?
Tomer Cohen: 我们在组织内部正式设立了全栈构建者这个头衔,我们正在逐渐将人们归入这个类别。
Lenny Rachitsky: 所以正在形成一套完整的职业阶梯。有一整套……好的。这比我想象的还要重大。那么你发现这些人主要来自哪里,比如产品、工程、设计?我想应该是混合的,但有没有最普遍的趋势?
Tomer Cohen: 是混合的。正在听节目的人,我建议只需回顾一下你的组织,想想谁能做到这一点,谁现在就能在这些职能之间灵活转换,无论是工程、设计、产品,甚至是 BD,你会发现已经相当多的人能够跨界了。
Lenny Rachitsky: 有趣。你认为有哪些职能在这方面做得特别成功吗?不是要偏袒谁,但我不知道。你有没有发现比如哪个更好?或者你也可以不特别指出任何一个。
Tomer Cohen: 没有,我认为这是一种关于如何做事的心智模型。我想如果要挑出一个可能最难学的手艺,我认为设计还有更多的工作要做,才能让设计智能体变得非常、非常出色。所以我认为,相比于其他角色学习设计的技能,设计师在学习他人技能时稍微有一点优势,反之则不然。但我老实说,这是一种思维方式。我见过设计师写代码,也见过产品经理做设计并且做得很好。这就是为什么我认为当你退后一步,思考你组织中的人以及谁能灵活转换时,你会发现他们出现在许多领域。我认为你在那里会发现的是他们有主观能动性,他们乐于接受新事物,他们很熟练,比如他们已经在构建新的体验,并且他们拥有那种想要变得更好的成长型思维,所以他们在学校学了什么或者加入公司时别人给他们贴了什么标签都无关紧要。
角色转换的激励与时机
Lenny Rachitsky: 我非常喜欢这其中的很多方面,因为现在是在不同产品角色之间转换的最简单时期,比以往任何时候都容易。设计转向产品经理,当然,或者只是转向这个新角色,它让转换变得如此容易,就像你说的,那个研究员变成了增长产品经理。
Tomer Cohen: 这可能是我给团队最大的建议兼动力,因为我告诉他们,归根结底……顺便说一句,这对我来说也是一样。我也以同样的方式思考。你的动机与组织对我们的要求是如此一致,对吧?因为我们需要你改变。我们想成为一个更敏捷、适应力更强、更有韧性的组织,能够应对变革的步伐,但对你自己的职业生涯而言,你也同样需要这些。你想站在如何构建事物的最前沿。所以,你自身职业发展所需与组织需要你做的事情之间,动机真的是一致的。因此,那种去放手做这件事的许可,对我来说,理想情况下是他们在做自己想做的事情时最大的助力,胜过其他一切。
落地转型的具体建议
Lenny Rachitsky: 也许对于受到启发并觉得“好的,这就是我们需要做的”的人,最后一个问题,对于那些刚刚踏上这条路的人,在自己的公司尝试这样做并取得成功,你有什么建议吗?
Tomer Cohen: 我想说的是,我会从如何为你带来这种结构的概念开始。我会思考你需要构建的平台、你想要引入的工具,然后我会在企业文化上花很多时间。我认为平台和工具再次成为先决条件,但并不充分,而文化层面真的非常重要。我会深入思考如何带领大家共同前进。所以关于我们的一个经验教训,如果现在重新做这个项目,我们可能会采用不同的做法,那就是有一段时间我与此事的核心团队紧密合作,这个核心的全栈构建团队负责构建所有这些材料,但组织其他人总是在问问题。“发生了什么?谁在做?有哪些工具?”回想起来,我们本可以在工作流中做更多的事情,直接向他们展示并让他们已经开始使用早期工具或者了解它,而不是在旁边搞一个小团队。
所以从小团队开始是可以的。我认为这真的很重要。但与此同时,确保整个过程的可见性是非常强大的力量。保持耐心并愿意投资。我总是举这个例子,我们总是举这样的例子,“哦,看看这家初创公司,他们在一周内就建成了这个。”是的,如果你从零开始,你现在确实可以在一周内构建出一个生活方式应用。这其实并不难。但是当你试图转型一个大型组织时,你需要对目标抱有不耐烦的态度,你必须有很高的抱负,但在如何将其变为现实以及你必须投资的关键事物上,要非常深思熟虑和有耐心。如果你不在你的平台上投资,我只是看不出这怎么会有一个成功的结果。如果你不投资为你定制工具,那么你只会从外部得到千篇一律的普通智能体。
所以要意识到这项投资,并确保你真正为其分配资源,这有点像经典的说法,愿意前期投资以便以后收获收益,而不是说,“嘿,为什么我没看到我们在一周内生产力翻倍?”情况不会是这样的。你可以在一些人身上看到这一点,但开始收集这些例子并开始真正思考转型才是关键。
Lenny Rachitsky: 这真的太酷了。我知道许多首席产品官、产品负责人以及各种领导者都在联系你,试图弄清楚你学到了哪些关于如何做到这一点的经验。所以我非常高兴我们能深入探讨所有这些事情。最后一个问题,在我们进入非常刺激的闪电问答之前,还有什么我们没有分享过的事情,你认为可能对听众有帮助,或者只是想再次强调的吗?
**Tomer Cohen:**无论你是在一个组织中等待你的领导者来推行这个,还是你是一个试图推行它的领导者,我都不会等待。我做的第一件事,回想起来我觉得非常有希望的是,我预先宣布了我们将会采用这种模式。我们从小范围开始,从小组开始,我们在构建工具,但这是我们将要攀登的高山,而且在很多方面,我们会把它做得很棒。我也宣布了这不仅仅是一个最终状态,而是一种持续的进步。并没有一个我们将要达到的状态,更多的是在不断地试图变得更好。而且在很多方面,为了竞争,你只希望在构建方式上比其他人更好,因为构建的版本每隔几年就会完全自我转型。
所以不要等待。真正专注于你正在取得的进展,与你的团队过度沟通,不仅是愿景,还有你正在取得的进展,几乎就像是在让自己负责。如果你是一个领导者,给你自己设定与你自己的团队分享的 KPI 或 OKR。如果你在组织内部,我想说,无论你的 CPO 或 CEO 是否在宣布这类项目,自己去做,或者加入一个这样做的组织,这样你就能处于未来构建方式的最前沿。
闪电问答
**Lenny Rachitsky:**Tomer,说到这里,我们到了非常刺激的闪电问答环节。我有五个问题要问你。准备好了吗?
**Tomer Cohen:**我准备好了。
**Lenny Rachitsky:**第一个问题,有哪些两三本书是你发现自己最经常推荐给别人的?
**Tomer Cohen:**我喜欢给出三本我真正喜欢的书。所以我当前的三本组合是,它们在主题上非常多样,所以如果它们不全是科技类的,我表示歉意。但第一本叫《Why Nations Fail》。这是我在十年前甚至更早读的一本书,它的作者去年刚获得了诺贝尔奖。它基本上讨论的是为什么有些国家成功而有些失败?而且这不是我们通常倾向的常规解释,比如,哦,是文化,是自然资源,是某种宗教。很多那些往往是人们拥有的直接借口。它大致分为两个阵营。是汲取性或包容性制度(extractive or inclusive institutions)?人们能否广泛参与并且机会共享,还是存在基本上理应从多数人那里汲取并给予少数人的制度。
所以这仅仅是一种令人难以置信的方式,去思考你如何构建一个国家。对于我们在 LinkedIn 来说,我们思考了很多关于机会的理念,所以也就包括你如何构建一个产品。而且这只是一个好方法,可以摆脱简单的解释,深入到真正让一个国家真正成功的原因。第二本书,叫《Outlive》。它真的是关于这个理念的,有点像作者 Peter Attia 谈论的医学 3.0(medicine 3.0)的理念,这其实是构建个性化医学的概念,我认为在 AI 的世界里,这在未来会变得不可思议。但这涉及所有那些,我们姑且称之为你应该为你的生活考虑的类别,这样你就可以尽可能地优化你的健康,它涵盖了一切,从健身到饮食再到你应该考虑的最大健康因素。但这是一本很棒的长书。然后最后——
**Lenny Rachitsky:**就在我身后书架上的那本。
**Tomer Cohen:**没错。
**Lenny Rachitsky:**在最上面。我想你其实是看不到的。
**Tomer Cohen:**然后最后,这也是很多年前出版的一本书,叫《The Beginning of Infinity》,我非常喜欢,作者是 Deutsche。这对我来说不是一本容易读的书,但我喜欢这个理念。事实上,特别是在产品中,我喜欢因果关系的理念,就像真正为事情为什么发生找到极好的解释,然后在此基础上构建你的下一次迭代。
这本书真正推进了解释的理念,只有一旦我们对事情的发生有了清晰的理解,然后我们才能在此基础上取得突破。但在我们达到明确的科学突破点之前,我们不会取得重大进展。但当你做到这一点时,真的几乎就像你可以在其之上取得无限的进步。
**Lenny Rachitsky:**Naval 总是在谈论最后一本书。我想我买了它,但只是读起来挺难的。
**Tomer Cohen:**它不是一本容易读的书,至少对我来说。它不是一本容易读的书,但它是一本非常有分量的书。
影视与播客推荐
**Lenny Rachitsky:**太棒了。有没有哪部你最近看过的最喜欢的电影或电视节目?
**Tomer Cohen:**我可以说个播客吗?
**Lenny Rachitsky:**当然可以。
**Tomer Cohen:**所以有一个播客,它是希伯来语的,叫 One Song,它拿一首通常是理想情况下很流行的歌曲,然后非常深入地探讨这首歌的起源和历史,我很喜欢它。我热爱音乐,它把歌曲剖析得非常好。它在赋予歌曲背后的故事以生命力方面也做得非常出色。所以对我来说,它回到了你原以为这首歌是关于某事的,但随后它会非常深入地探究歌曲背后的人物,有时是选择的词语,或者是歌词如何与音乐本身相匹配,我真的非常享受这个。
**Lenny Rachitsky:**我相信有一个叫 Song Exploder 的播客,是一个类似的概念,不是希伯来语的,是英语的,如果你喜欢那个的话,我会向大家推荐这个。
**Tomer Cohen:**太棒了。
近期最爱产品
**Lenny Rachitsky:**有没有你最近发现的、你非常喜欢的某个产品?可以是一个应用程序,可以是一些衣服,也可以是一个厨房小工具,那一类的小工具。
**Tomer Cohen:**它可以是一个我想要拥有的产品吗,我认为这其实很容易做到?
**Lenny Rachitsky:**我喜欢这个。这就是产品思维 101,以及你想要看到的愿景。
**Tomer Cohen:**所以现在在我的车里,有内置的 Alexa,这很棒,因为孩子们可以整天点歌,这简直就是车厢里的一整场表演。但是我一直喜欢做的一件事,这我已经做了两年多了,就是我进去并进入语音模式。
**Lenny Rachitsky:**ChatGPT。
**Tomer Cohen:**是的,ChatGPT,然后只是进行对话,但这有摩擦力。我很想在方向盘上有一个按钮,可以唤起坐在副驾驶座上我旁边的我的 AI 朋友,我只是觉得那将会是一种……我其实认为它会 [听不清] 人们的出行。仅仅是那个动作,那种摩擦的消除就会为我改变体验。
**Lenny Rachitsky:**说到这点,我最近发现特斯拉现在其实已经做到这一点了。如果你按住右侧方向盘,Grok 就会出现,你可以和 Grok 说话。所以它来了。AI 已经降临了。是的。我只是不小心按到了,然后它就说,“好的,酷。”
**Tomer Cohen:**太好了。所以对我来说,如果有任何来自 Rivian 的人在听,请把这个带到车里来。
**Lenny Rachitsky:**Rivian 落后了。是的。而且你必须使用 Grok。如果你能切换到不同的 AI 那会很酷,因为它有一种性格。只要给我信息就行。我不需要你笑和给我讲笑话。
**Tomer Cohen:**你之前需要花点时间适应它吗,还是它有任何来自……的记忆?你有带入任何记忆给它吗?
**Lenny Rachitsky:**有一个未登录版本,然后你可以直接登录,它就会连接到你的账户。是的,它极其酷。没有人在谈论这个。这很疯狂,因为我不知道他们是否已经全面推出了它,但它就是出现了。
**Tomer Cohen:**你在车里经常和它说话吗?
**Lenny Rachitsky:**老实说,我没有那么频繁地使用它,但我应该多用用。我妻子就是不太喜欢 Grok。我觉得 Grok 这个品牌有一种很特定的调性。所以她会说,“别在我旁边跟 Grok 说话。”
**Tomer Cohen:**我很喜欢语音模式,所以我一直在用。
**Lenny Rachitsky:**是的,我也很喜欢语音模式。只是它打断我的次数太多了。这就是问题所在,对吧?它就是会停下来。
**Tomer Cohen:**顺便说一下,你可以设置一下。你基本上可以说,“嘿,让我把话说完。”
**Lenny Rachitsky:**我现在知道了。今天真是学到了很多。好吧,还有最后两个问题。你有没有一句在工作和生活中经常觉得很有用的人生座右铭?
人生座右铭与成长心态
**Tomer Cohen:**我想上次谈到这个时,我最常引用的是“我可能错了,但我并不困惑”,尽管我现在不怎么说了。但我认为我真正喜欢的是,成长型思维在我们家堪称第二信仰。其中我很喜欢一句话,就是“成为比存在更好”,我觉得这和全栈构建者模型有一点联系,也就是你始终处于进步模式、迭代模式。这不在于达到某种状态,而在于旅程和过程。这才是你应该热爱的地方。它关乎持续地成长和进化,没有其中的负面情绪,也没有那种错失恐惧症的感觉。它就是这种持续不断的东西。如果一年后我回头看,我会想,我成长了多少?我掌握了多少知识?有哪些技能可以再次发挥作用?我在哪些方面变得更好了?我能否感觉到 2026 版的 Tomer 相比 2025 版的进步?其中的增量是什么?我挺喜欢这种思考方式。
**Lenny Rachitsky:**这完美地过渡到了我们的最后一个问题。等这期节目播出时,你在待了 14 年后离开 LinkedIn 将不再是个秘密。这是一段传奇的旅程。你在被收购前很久就加入了,你帮助他们完成了整合。就像 14 年前人们对 LinkedIn 的看法与今天截然不同。现在在那里的感觉其实非常有趣和有意思,而不是人们长期以来对 LinkedIn 的那种感受。所以我想问,你现在感觉如何,下一步是什么?我想会有很多人给你打电话,但你有什么计划?
离职 LinkedIn 与未来规划
**Tomer Cohen:**是的,所以我感到很自豪。在 LinkedIn 是一段难以置信的旅程。我第一次深入了解 LinkedIn 是在 2008 年搬到硅谷后,去斯坦福听了一场关于社交网络的讲座,Reid 也在那里,他谈到了在线专业社区的力量。我对这个话题非常痴迷,认为这是一个不可思议的愿景,当时并没有加入的计划,实际上之后还创办了自己的公司。但碰巧的是,几年后我加入了这里,只是觉得这个使命太棒了。所以在很多方面,它都与我的目标相契合,在这里真是一段不可思议的旅程。
同时我也非常感恩。最近我刚和公司分享了这些。我开始从这里的经历中汲取经验。其中很多来自于艰难的处境。在 LinkedIn 我们经历过很多艰难的处境、艰难的抉择和熬夜,但你从中学到了太多,我只是无比感激。而且我很兴奋。我很兴奋。我有拥抱变化的倾向。我倾向于将自己置于一个能学到最多、学到很多东西的位置。现在是构建的绝佳时机,所以我非常兴奋能去思考新的问题集和新的领域,在这些地方深入探索,并投入接下来的十年。
**Lenny Rachitsky:**我想你要花很长时间才能不再觉得自己在为 LinkedIn 工作,才能忘记这些年来你一直担忧的所有事情。
**Tomer Cohen:**在你花了这么长时间构建某样东西之后,我觉得我们之前在某个时候聊过,我认为构建者最好的特质之一就是对他们正在构建的东西变得非常充满激情。真正在乎。不是在乎这份工作,而是真正在乎这个产品。当有人抱怨时你会感到痛苦,你会有这种持续的不满,对我来说,这就好比抚养一个孩子的概念。所以是的,这很难。将会很难。我会永远把 LinkedIn 视为我帮助长大的孩子之一。
**Lenny Rachitsky:**好吧,我很期待将来有一天,等你弄清楚下一步想做什么,或者开始做任何你想做的事情时,能再请你来做客。我很高兴能以这个借口来认识你。Tomer,非常感谢你的到来。
**Tomer Cohen:**很高兴来到这里。谢谢,Lenny。
**Lenny Rachitsky:**大家再见。非常感谢大家的收听。如果你觉得这期内容有价值,可以在 Apple Podcasts、Spotify 或你最喜欢的播客应用上订阅本节目。另外,请考虑给我们打分或留下评论,这真的能帮助其他听众找到这个播客。你可以在 lennyspodcast.com 找到所有往期节目或了解更多关于本节目的信息。我们下期再见。
术语表
| 原文 | 中文 |
|---|---|
| agents | 智能体 |
| APM | 助理产品经理 |
| Associate Product Builder Program | 助理产品构建者项目 |
| BD | 业务发展 |
| design systems | 设计系统 |
| Deutsche | Deutsche |
| extractive or inclusive institutions | 汲取性或包容性制度 |
| first principles | 第一性原理 |
| FOMO | 错失恐惧症 |
| full stack builder model | 全栈构建者模型 |
| Full Stack Builder Program | 全栈构建者项目 |
| GA | 通用可用性 |
| Lenny Rachitsky | Lenny Rachitsky |
| medicine 3.0 | 医学 3.0 |
| meritocracy | 精英体制 |
| Naval | Naval |
| Peter Attia | Peter Attia |
| PR | 拉取请求 |
| Reid | Reid |
| Rivian | Rivian |
| Shira Gasarch | Shira Gasarch |
| Tomer Cohen | Tomer Cohen |
| trust agent | 信任智能体 |
此文档由 AI 分片翻译(translate_long_document)