# Built on a Crisis: Jeff Wang on Winning Enterprise AI Coding with Windsurf

## Метаданные

- **Канал:** ProductLed
- **YouTube:** https://www.youtube.com/watch?v=SY8d2oBXizE
- **Дата:** 24.04.2026
- **Длительность:** 35:34
- **Просмотры:** 33

## Описание

When Jeff Wang stepped into the CEO role at Windsurf, it was not part of some long-term succession plan. It happened in the middle of a full-blown crisis.

In this episode of the ProductLed Podcast, Wes Bush and Esben Friis-Jensen sit down with Jeff to unpack the wild chain of events that followed the collapsed OpenAI acquisition, the founders leaving for Google, and the intense 72-hour window Jeff had to help save the company and protect 250 jobs. He shares how Windsurf navigated that moment, how the Cognition deal came together, and what it has been like leading one of the most closely watched teams in AI coding ever since.

Jeff also gets into what made Windsurf so strategically valuable in the first place, from shipping early breakthroughs in autocomplete, chat, context engineering, and agent workflows, to building one of the first generally available coding agents on the market. Beyond the origin story, the conversation goes deep on go-to-market strategy, why free products worked early on, how token economics changed the game, and why enterprise AI adoption takes far more than handing teams a tool.

They also explore Windsurf 2.0, the shift toward managing multiple agents at once, how Jeff uses AI in his own CEO workflows, and why founders need to obsess over painful problems, customer conversations, and product-market fit instead of flashy demos.

Key Highlights:
00:00 - The 72-Hour Crisis That Changed Everything
Jeff shares the short version of the OpenAI, Google, and Cognition saga, and what it was like stepping into the CEO role during a company-defining emergency.
01:40 - Why Big Tech Wanted the Windsurf Team
A look at the execution speed, product breakthroughs, and agent innovations that made Windsurf one of the most valuable teams in AI coding.
04:10 - The Future of Coding Is Multi-Agent
Jeff explains why developers are moving from one-on-one AI assistance to managing many agents at once, and how Windsurf 2.0 is built for that shift.
08:54 - How Free Became Their Growth Wedge
From free autocomplete to on-prem enterprise deals, Jeff walks through Windsurf’s early PLG motion and how it created awareness and pipeline.
13:10 - The Hard Truth About AI Pricing
A candid discussion on token costs, self-serve subsidies, pricing pressure, and why raising prices can reveal whether you truly have product-market fit.
16:13 - Why Enterprise AI Sales Are Top-Down
Jeff shares how Windsurf sells into large companies by focusing on transformation, adoption, security, and measurable outcomes instead of seat counts.
20:51 - What It Takes to Drive Real AI Adoption
Why playbooks, training, and solving a meaningful first use case matter more than just rolling out a shiny new tool to an engineering team.
24:40 - Jeff’s AI Workflows as CEO
Jeff reveals how he uses AI and custom playbooks for go-to-market research, outreach preparation, and spotting product trends before opening dashboards.
32:32 - Jeff’s Advice for Every Product Founder
Build around painful problems, talk to hundreds of prospects, and learn to enjoy rejection because that is often where the real insight comes from.

Resources:
🚀 Windsurf: https://windsurf.com/
💼 Connect with Jeff Wang on LinkedIn:   / jefflinwang  
💼 Connect with Wes Bush on LinkedIn:   / wesbush  
💼 Connect with Esben Friis-Jensen on LinkedIn:   / esbenfriisjensen  
📚 Read the interview summary: https://productled.com/blog/how-jeff-...
🎧 Listen on Spotify: https://open.spotify.com/episode/4w70...
🎧 Listen on Apple Music: https://podcasts.apple.com/us/podcast...
🧠 Sign up for the ProductLed Newsletter: https://www.productled.com/newsletter

## Содержание

### [0:00](https://www.youtube.com/watch?v=SY8d2oBXizE) The 72-Hour Crisis That Changed Everything

I think some CIOS might think that rolling out a dev tool is all you need to do, but then they're going to look at their adoption metrics and really see like, oh, actually, it's kind of weird that only maybe like 15 20% of the people are actually using these tools. And why is that, right? If everybody has AI, then nobody has AI. A lot of these GTM AI tools rolled out and then I got spam with like hundreds of emails. I just like pretty much like banned all those. I blocked all those emails. So, you need humans to really figure out how to differentiate. You have to feel that it's fun to get a customer to say no to you. And then it has to be like a game to try to flip that to a yes. If you don't think it's fun to get rejected, you should not be a founder. Welcome everybody to the productled podcast. With me I have my co-host Esben who is the entrepreneur and resident at ProductLed who is also the co-founder of user flow and cobalt. And our guest today is Jeff Wang, who became the CEO at Windsurf, not by choice, but by crisis. And so when the founders left for Google after an open AI deal collapsed, you had literally 72 hours to save the company and 250 jobs, which is crazy. So you pulled it off, Jeff, landing an acquisition as well with cognition and making sure every employee got paid out of this deal. And today you are leading one of the most talked about teams in AI coding. So Jeff, it's awesome to have you here. Thanks so much for coming on. Yeah, thanks for having me by the way. — Awesome. So, could you maybe share a little bit just a quick TLDDR on the story from like the Open AI deal to what happened with Google because I was reading about it this morning in preparation for this and I was just like this is bonkers. Like it's crazy how fast everything happens. But I'd love to hear your take on it.

### [1:40](https://www.youtube.com/watch?v=SY8d2oBXizE&t=100s) Why Big Tech Wanted the Windsurf Team

— Yeah, it kind of was like the who's who in the tech world. I think after the OpenAI acquisition news leaked, Enthropic actually quickly like moved to shut us down and after that Microsoft and OpenAI started arguing about the IP rights. I'm just I'm only reporting what I can in the public that was released and then of course like Google stepped in. You pretty much have like all the major tech companies that are being talked about all kind of like arguing or fighting for this uh this team. Um and then eventually of course like the Google team or the part of the team went to Google and then part of it stay behind and then rest went to uh cognition and that's kind of the TLDDR but uh it was a lot of stress and a lot of uh yeah a lot of moments that I can't talk about but like a lot of crazy stuff even that's publicly talked about. I mean, this is kind of like a dream for a lot of founders, too. You're like, "Hey, everybody's fighting to like acquire us or give us lots of money. " But what was so key and critical about the Windsurf team that like all of the big tech companies were fighting to whether it's Aquire or acquire the company to get access to this talent and IP and resources you built? — Well, a lot of it was execution velocity and just being correct on a lot of the uh suspicions. So if you look at Kodium which was a product before windsurf it was the uh first product to actually release chatbt integration. So you had like your autocomplete and you had the chat interface as well and then it was the first company to do a lot of things like context engineering like getting parts of the codebase into the prompt so that you could get better results and then there was things like pinning context and getting different deployments to enterprise. So we were like the first to move on a lot of these features that everybody else has. And obviously the most uh biggest breakthrough was the agentic piece. So you're taking the entire conversation putting it back in until you finished the plan. You've executed on the plan and that like was the first generally available agent on the market and uh which was a huge game changer of course to everybody else in uh kind of copying that uh over time. And then if you look at it at cognition now, Devon was the other like truly agentic um piece of software that was on the market as well. So both Insurf and Devon were like first of its kind novel ideas that the whole world is kind of trying to copy right now. — Oh, definitely. And so now is your first term as CEO of Windsurf. How's the first period been for you? Because I'm sure there was like a lot of change initially and then have you kind of settled into your groove now? — Every week is very different. Um, we act as like pretty much one company now.

### [4:10](https://www.youtube.com/watch?v=SY8d2oBXizE&t=250s) The Future of Coding Is Multi-Agent

So, the integration was a tough piece as well. But I would just say the job of a CEO or like anybody at the top is it could be very random week over week. Um, lots of travel as well. I think I've been in like a dozen different countries now in the past like few months. Um, and it's just really whatever is the most important use of time just try to go and spend time there. — Definitely. And how do you differentiate like wind surf compared to all the other kind of platforms because who is would you say is like your top maybe two to three competitors that are on the market? — Yeah, right now the most common that we run into in enterprise it's probably cloud code and maybe even codeex now but basically the whole world is transforming to this agent only this endto-end agent and like being generally um available. So basically like you can ask a very general question and it's able to understand how to solve it and go across all your systems in your company to make it happen. And what happened after that is like you know people used to sit in front of an ID and they go one-on-one with an agent but now that the agent can do the all these things end to end like verify the code and even run the code and test it on a UI. People are just sitting around doing nothing for like hours now. So, so what they're doing is they're running another agent and another agent because if they're taking several hours, it's not uncommon now to see people have like 10 agents on the screen, which is actually what Winerf 2. 0 is all about. Um, we want to make it easy for folks to manage like dozens of agents at once. — Awesome. One just diving a bit deeper into that competition piece. So, one part is of course Devon, right? Devon you I guess used to be a competitor. Now you're kind of looking how can the two u two products kind of cooperate or work together. So that's one question and then you mentioned claude code but I also know that aren't you using some of their models as well for windsurf um and how does that work? So, how do you think of all of this? Like, because I think you I read somewhere that you kind of see yourself as neutral in the whole model game, but how long can you stay neutral if there are all these players also doing uh the same things as you're doing, but they're also doing models? — Yeah, I think back So, the first question is Devon, right? So, Devon is actually not a competitor even back in the day. It was very rare where we went head-to-head against a Devon pilot. And the reason why is because it is a very different surface area. Uh you are running Devon in the cloud and Devon is going like end to end whereas Windsurf you're kind of like sitting in front of the computer and working with the agent to together. And the combination of both these form factors is actually very interesting because if you take any software engineer now it really is do they work with an IDE, a CLI or do they work with an a remote agent, right? And a remote agent is very important because if you want to scale agents across a company, you want a very kind of defined kind of a system where everybody has the same access, right? Like if you think about openclaw, you can't just roll open claw to everybody because you're giving everybody access to everything. But with Devon, you're giving everyone access to this VM that has access to the right amount of data and access to the repositories documentation. So that is really good to scale because you know exactly what everybody needs and you could spin up like thousands of VMs uh like you can have agents running everywhere all doing very important things and that is very hard to do even internally if you build it yourself or uh you're using um kind of some of the other solutions that are trying to do the sandbox. And then to your second question about kind of model being model agnostic like there's a couple things here. One is that you know users on Windsurf can choose the model that they want and we even we might be the only company actually that has our own model that we serve for free and of course we don't know how long they'll be free but we're trying to make it free as much as we can and the reason why is because people are blowing up their token budgets and they need alternatives now. um they have no choice but to kind of use some of the features that we have for example like the free models that no one else is offering right um and then the other part is like how can we compete like if you look at Devon is using a whole bunch of models under the hood and there are different things that the models are good at and they're not good at and we have a very like intense eval team here and we choose the right model for the use case so I think that's like something that's very important for someone that might go all in on one model provider you actually you actually might put yourself behind in some cases for some use cases. — Definitely. And so when you joined, what was like where was the company at? Because I know when you got like some of the initial offers uh you were at about like 84 million AR R and so just curious like what was that the first journey when you started like where was the company at and then what were some of the things you did to grow so quickly?

### [8:54](https://www.youtube.com/watch?v=SY8d2oBXizE&t=534s) How Free Became Their Growth Wedge

quickly? — So I was the first Kodium hireer I think the first Kodium email and product was not making revenue yet. uh and we actually debated like what were we ever going to create a SAS component uh that was generating revenue. The only thing we did was we released a free SAS product and by the way we operated our own GPUs and hosted our own models so we can bring the cost down to server it for free and we just gave out autocomplete and eventually chat uh all for free and the reason why is because we wanted to just get people to look us up on the internet and be like oh what's what is Kodium? I think to look you up initially was a extension in like VS code. — That's right. We had extensions across all the a lot of the IDs I should say almost all the ids and um and then that would all connect to this same uh GPU cluster where people could run the autocomplete and back then you would there was only copilot and you would either pay like $15 a month or something uh back then or you could just download uh Kodium for free and then you look us up and you realize there's an enterprise plan but the enterprise plan at the time was only onrem like we would serve this same box to your organiz ganization and that was kind of our strategy uh pre-revenue to get our first customers because you always want to choose kind of this um this vertical which is not as competitive, right? Because if we go headto-head with co-pilot, it's like you guys are like less than like a dozen people. Like how can you how can we trust over like our code base and stuff, right? Um so that was kind of the strategy is like go after this very high demand low competition arena and of course it's very high like a lot of manpower is needed to deploy these a lot of engineering effort is needed but that is like what our team was good at right so we were able to get our first revenue from that method — okay yeah so if I understand correctly you really just stood out with the autocomplete feature initially kind of built your wedge got your I think it was like a million plus users with Or was it or how many users did you have? — Couple hundred grand. Yeah, a couple hundred grand. A couple hundred thousand I should say. 100 grand. — Definitely. Hopefully more than that eventually. — You're used to being in dollars. — Yeah. — The CEO mindset now. — Maybe. Also, I'm like recovering from this like very mega illness like I said in the beginning. So, maybe my brain scrambled. I don't know. — Yeah, maybe. No. — Cool. So, you always had that free motion. Now, um, why was that super important for your market? Because, uh, like if you look at a lot of developer tools, we see this all the time. It's like they usually are all productled, but I'm curious like why did your team kind of decide to have that initially free? I mean, it was probably the only way we could get users. And if you look at the marketing campaigns we did, it was all about how it was free. It was not about like we have this enterprise product and it's going to like be very uh, distinguished from anything else on the market. It was all about this is free. It's a no-brainer to use it. driving conversions to downloads and as long as you downloaded it at least we would build awareness and if you go to enterprises they would probably have somebody that uses it and then they could be like oh I've heard of that company if they have an on-prem product I know how it behaves and we should go and buy them right so that was kind of the logic behind that — okay and so for those thinking that are product life founders that just are like oh yeah great that's the only answer is free product then we're good but that doesn't solve the business model side of things so you uh went from like selling, you know, $1,000 per year to like millions. So, what changed when it came to how do you actually get or capture a lot of that value you create with that free model? Uh what were some of the first steps you took because you started this product from like zero basically from a revenue model? — Well, you have to look at pipeline, right? So, if you are rolling out a free product, you need to have a pipeline that has high value. And when you think about it, the on-prem product was extremely high margin as well. So you would roll out something that the customer pays the GPUs for and they roll out and they pay the license to you. So we were actually making like a lot of money on the product that we were rolling to enterprise and then our costs for uh PLG were actually fairly low in those days um because um we were able to control the cluster and infra as well. Now when you look at how

### [13:10](https://www.youtube.com/watch?v=SY8d2oBXizE&t=790s) The Hard Truth About AI Pricing

when Windsurf came out, this is another story because the token costs are way higher like you actually have to subsidize the tokens for the self-s serve model and that is a different story to then when you get to enterprise you really need to like optimize for the enterprise licenses to be to cover the cost of self-s serve and this is by the way probably the number one problem I would see founders trying to enter the arena today. I feel like we might have broken the cost in terms of how to run these models and now it's like the norm to spend like even like tens of thousands of dollars per user per month and that is like not going to be easy for a founder to do if they're trying to roll out a product that is as capable, right? — Yeah. Which kind of gets to how do you make that work? So when you're subsidizing these free users uh and you're like hey we got to make this balancing act of like okay we got to make enough on the uh revenue of the enterprise and if you have prousers as well but basically every founder I've talked to recently that has a AI native product they are to some degree subsidizing like quite substantially these free users and that's why you do see a lot of them raising money because they're like this thing is not that sustainable. So I'm curious how you handle it now and you're planning on handling it in the future too. — Yeah. Well, I think for us it's a little different. We've kind of established that AI coding is very valuable. So it's at the point where like if we subsidize then people are just using us as like arbitrage. Like they're just getting cheaper tokens from one location to the other which is by the way we had to repice all of our self-s serve products because people it was like growing too much in the worst way possible. It was growing because people were trying to subsidize their own token use. And if that is the case, that means that is not PLG anymore, right? That is more like you're just giving money away. And I think if you have product market fib, you can raise your pricing and you your revenue should not be affected. Uh or meaning it should go up, right? And if you raise your pricing and everybody goes to another provider, that's a bad sign. That means you no you don't have product market fib you were never really in PLG to begin with because you were just people were just using you for token arbitrage. Um so that's something that's very important because that means a founder should really strive for something that has product market fib that is differentiated that is that you can compete to your target audience right for us it's enterprise so we are able to compete in enterprise because we have a lot of other factors for example we drive outcomes we don't drive dev tools right like we go to our enterprises and and understand what it is they're trying to build and like how much that costs and can we do it for a lot cheaper and faster right so that is very different product uh mindset than you would if you were rolling out a new new new tool, getting free users, subsidizing it and then trying to convert into payments. That is a very different motion actually if you think about it. — If you speak a lot about this with enterprises and developers, right? So what is your funnel today? Is everything coming from a developer who goes through the free motion and then kind of recommends it inside the enterprise or are you doing what are you doing actively to uh sell to these enterprises basically?

### [16:13](https://www.youtube.com/watch?v=SY8d2oBXizE&t=973s) Why Enterprise AI Sales Are Top-Down

basically? — Yeah. So I think most of the revenue is driven from the top down meaning we meet with some of the leaders of the largest companies try to really understand their pain points and where they're trying to save money and drive AI transformation but there is a subset where you can make it easier for people to get your their hands on the tools. So for example, we just rolled out Windsurf 2. 0 last week. You can use the Windsurf login now to log into Devon. So you can actually try out the other platform, other parts of the platform by using the tools that were easy to deploy earlier, right? Because Devon, I'll admit, is like you need to connect it to a code for it to have a lot of value. And that is actually kind of hard to drive self-s serve usage without something like having Windsor 2. 0, installing it locally, driving a lot of value, and then just logging into Devon as a result, right? So there's some motions where you have existing easier entry points and then kind of getting people to your other products as like a PLG motion. And then of course we're trying to lower the barrier of entry of course in general to the seller for some of these products as well. How — how do you think about it? Because when you look at your market, the coding market and so on, right? Uh almost all your competitors are very PLG and a lot of the go to market is driven through word of mouth and bottomup, right? So a lot of their customers are single person shops and stuff like that, right? But still paying money, right? But it sounds like you have a more enterprise approach. How do you balance it? Right? Because you also want to be a popular tool and get the word of mouth, but you also want to earn money. How do you find that balance? And it sounds like you more gone towards the enterprise side of things where some of the others maybe are just trying to get a lot of word of mouth in the lower end of the market. — Understood. There is an equilibrium, right? There there's an amount of money that you're willing to spend on we call it like marketing in instead of EG. Um and then the amount you're willing to see how much money you get back from it, right? And you could obviously in the extreme sense you can give everything away for free and see how much revenue you get or you can charge a lot of money for it up front and see how much money you get. So it's it obviously it lies somewhere in between and we've been experimenting with different price points and trying to figure out what where that point is. And again recently we've just had to raise the prices because people were not the right kind of user. If you're only using the product because it's cheaper, that is not the right kind of user, especially if you're losing money. And I think that is maybe a hard truth to swallow for a lot of founders. — And do you maybe just one add-on question to that. So if you go more top down and stuff then I would guess what are your arguments then? Is it security? What is it that you bring to the table that the others can't because you're targeting this other ICP? You can say — yeah I mean a lot of the work we drive is very difficult to implement. So a lot of the AI transformation is not easy. I think some CIOS might think that rolling out a dev tool is all you need to do, but then they're going to look at their adoption metrics and really see like, oh, actually it's kind of weird that only maybe like 15 20% of the people are actually using these tools. And why is that, right? Well, a lot of it is because we have to go in and drive uh adoption and training. And another part is we have to go in and drive kind of the security and kind of getting access to the tough uh like very sensitive code bases or sensitive databases or connecting the things that really drive a lot of value and a lot of companies are just not willing to do that or they're still trying to scale the team to do that and that is a huge difference when we go to talk to customers. It is like working with an actual someone that has a partnership to get the outcomes that you're trying to drive versus, hey, I'm the sales guy and I'm just trying to get as many licenses as I can from you. That is like that is not what we how we operate. We go in and figure out what the biggest problems are. And that again very different take uh when you chat with our team versus another team. And if there's a founder, by the way, trying to break into the industry, they need to understand this. A lot of the initial conversations was just us talking to like hundreds of customers trying to figure out why they're not buying us and then really understanding like oh they actually need a success out of this. This is not just like do I make a decision on how many seats to buy. It is what is it that they're trying to accomplish and try to make sure that they get that. — Yeah, I think it's worth kind of double clicking on you know the differences between a blue ocean and a red ocean. Like I know as Vinnie you started user flow it was like in the product adoption space it's uh at that time it was like okay great I know what product adoption is I just not the best fastest tool whereas what you're doing at Windsorf 2 it's like okay there's a lot of education that you have to do it's not a complete blue ocean in the sense like there's not no competitors but the market's so big you still have to do a ton of education on not just here's the

### [20:51](https://www.youtube.com/watch?v=SY8d2oBXizE&t=1251s) What It Takes to Drive Real AI Adoption

product here's how to use it but there's skills they need there's knowledge they need to actually fully utilize this and get the rest of their team to adopt it as a new behavior as well. So, what have you found has helped your team to really transform companies a lot faster than just like, hey, here's the tool. Go ahead, use it. Because clearly if they're only getting 10 to 20% of people using it, it's not going to get them to that successful outcome or transformation. — Yeah. I mean, one of the biggest things is finding what's the most important project that your tools very uniquely capable of solving. And for us, uh, for example, if you're, if you're a large institution that has a lot of legacy code, there's probably a lot of migrations and a lot of upgrades documentation, a lot of things that need to be done that has that you can actually just do right off the bat in the pilot, right? And um, that's one aspect of it. The other aspect of course though is the industry moves really fast and people a lot of people don't even know the difference between like sonnet and opus or like codeex and or like anthropic and openai and gemini. Like they have no idea, right? they're opening the tool and they're just like selecting the default and using it. That was a big surprise when we realized like globally people just have no idea what's going on in in Silicon Valley. We know everything that's going on. We see the changes and we're up to date on the trends, but no, nobody it seems like globally like very few folks and percentage- wise are really up to date. So we have to be the ones that drive the knowledge as well like keep them up to date and enable them to understand what is happening in the space because like I said we just rolled out Windsor 2. 0 know it's completely new form factor right and how do you even like assume everyone's just going to use this form factor you don't you cannot assume that right so we have to go to these enterprises and start driving that behavior right this behavior of even agents doing this end work um is very hard to drive um for someone that has no idea what AI is and then the second thing I want to point out is we have playbooks so we have things that are repeatable that you can drive so you go to the you set up these playbooks and then all they need to do is call these playbooks to execute some of the type of work that agent agents can do and that is actually very useful as well. when we go there and build these playbooks with our customers — and maybe even though you're uh maybe doing more top down today, you still have that original PLG culture and I think that helps you a bit when you're introducing these like new things. If you still have that culture that you can focus on things like it it's not only about selling it to the management, you also have to get the actual developers to use it and use it in a smart way, right? It's never either or when you're a PLG enterprise business, you're still using those PLG roots or something good. Is that also how you see it or how do you think about — I mean there's a lot of things that are still valuable from PLG like one of them is if you um want to experiment with features. So like we sometimes roll out features only to a subset of the users to see if they're adopting them or maybe like it breaks and like we have to figure out why or like they're using it improperly. So there's a lot of product feedback you can still get from like the self-s serve/PLG user base and um a lot of the feedback comes directly from them. If we go to enterprises with that feature and they complain it is like too late almost, right? You don't you like you don't want to spend your time fielding complaints from all your enterprise customers. You'd rather do it from the self-s serve users first and then get it right and then launch it to enterprise. — Nice. I love that. And maybe switching gears a little bit here too. I'm curious, how do you use AI to be an effective CEO? Like what are some of the things you found have been really effective for you to whether it's save time, keep tabs on just how the business is doing? But I'm curious if you got like any specific workflows or things you've created to just yeah be more effective at what you do there. So just

### [24:40](https://www.youtube.com/watch?v=SY8d2oBXizE&t=1480s) Jeff’s AI Workflows as CEO

last week actually I created a few playbooks. One of them is mostly go to market focus. So but it's like I want to be able to name a customer and generate an org chart for me, right? And this is plugging into um the XA MCP server. It's they they've indexed the web and they have LinkedIn as part of it and then basically Devon will just build me this like this chart of like who's reporting to who um at the high level and that workflow I can just give to everybody in the company now. Um it's just like something very simple, right? The other is a workflow I'm going to go to Google Cloud Next uh this week and basically all you need to do is call Devon. I believe I called it ghost note run the ghost note playbook and under my name and then it'll write it'll actually go retrieve who's relevant in that company get their email if you name an event it could guess if they're going to be at the event or not again this part is like you have to assume the agent's going to hallucinate a little bit and it will write it will write the note in my voice um and then I'm giving this to my sales teams so that they can send that to me uh like they can give me the temp like here Jeff can you just click this button to send the email. Um, so I'm still putting a human in the loop, right? I don't want this to send a thousand emails on my behalf. I still want the sales team to like go vet like, is this a good note and is this the right person to send to before I before it comes to me? And yeah, and then I just go and I send a bunch of notes. And I hope there's not a lot of other C-level people watching this and being like, wait, that's not coming from Jeff. That is just another example of a tool. Again, it's it is not like coding related, but it is I mean, behind the scenes, a lot of code is being written, but that is like a very useful tool that probably would have taken a lot of effort from the sales team, right? Got it. Yeah. And is there anything else outside of the sales side or just like understanding the business overall that you found has been helpful whether it's like a daily schedule task or weekly thing or something like that too that's really helped you? Yeah. So, a lot of the like token usage, um, product trends, for example, we did a big, uh, shift on, you know, Winer 2. 0, we did the pricing change. I I'm always asking the agents like, what is going on? Uh, what was the behavior? Did we do we just like lose a bunch of customers, right? Um, and it's funny because I kind of do that first before I even look at dashboards. So, I'm kind of like relying on more granular insights from the agents than looking at like the actual raw uh outputs of the dashboards, which is I think what a lot of other people are doing in the company too. Um I'm not exactly sure if that's a good thing by the way because the outcomes might be different from um session to session. Um but that is the behavior that's like very normal which is you ask the AI first and then you could like look at the data later if you want. So in this world where you can do almost everything with AI, Jeff, how do you actually build an organization because like do you need a sales team? marketing team? Because you know you can just start asking the AI to do some of the stuff they used to do. So what kind of profiles do you really need in a AI first organization? And maybe the bigger question for you is the role of the developer, right? Which is both somebody who works at your company but it's also your customer. Who's a developer today? like if everybody can code, who's actually a developer? What kind of role is that? — I'll address the first question first, which is I think actually more than ever you need marketing and sales folks because you first of all sales, you need people that go and meet people in person and build that trust, right? I don't think the world's ready yet to completely trust AI to do an entire transaction in and to be like, hey, like here's the features I need. AI, tell me to buy it or not, right? I think people need to have still have that human element. And for marketing, if everybody has the same access to a marketing tool, nobody has access, right? So like a lot of these GTM AI tools rolled out and then I got spam with like hundreds of emails, I just like pretty much like banned all those. I blocked all those emails. I can't I can't keep track of all these emails in my inbox from all the spam, right? So I think that's one thing is like if everybody has AI, then nobody has AI, right? For the GTM and marketing example. So you need humans to really figure out how to differentiate. And maybe you could use AI to differentiate, but at scale it becomes the same again. Right? For engineering, I think in terms of people that hire, I mean, you need people that are like still very curious that understand how to link systems together and understand the architecture of the entire stacks. And like we I I will say like our engineering team is way more efficient. So for example, we've closed 700% like so 7x more PRs in the last 6 months, but we our engineering team has only grown like 10%. So what I think is going to happen is people are just going to be doing a lot more, but you still need engineers. Like we're still actively trying to hire as many engineers as possible. We just I think the bar at cognition is just very high. Um but we I don't think there's a world where a startup is going to say like you know what like we should just cut half people. That is not going to happen because we need to move as fast as possible. Now when you have a large company then the the math might change. the a large company might say like oh we just don't have enough things to work on which is really weird for me to hear you know from a larger company but that is where like they might either hire less or even if you're not AI native like for example if you're AI native you're probably more than 10x more productive than someone that is not AI native then those non-AI native people are in danger because you're operating at less than 10% of uh productivity than everybody else right so that is something that I think people should be aware of is like this curiosity, this desire to use these tools is actually is helping your job. It's not it's not making it worse. Like you don't want to be left behind. — And so going back to your uh crisis week where it was like so much has changed in a few days. I was reading some of the past podcasts and stuff you were on and you said you lost like I think 8 pounds that one week. In hindsight, like what would have been like the self-management system you wish you had in place before the crisis hit? Like going back 2020 vision? — Wait, what do you mean by self-management? — Like what would you have done differently to either manage that situation better, whether it's like take care of yourself, your team better, or like prepare mentally for something like that? Because whenever any leader listening to this like hits a crisis, there is things that are like, "Yeah, I just drop that. " But would you have done anything differently? I'm curious. — I think the the reason why and by the way, I like I feel like I'm uh getting sick and traveling a lot now and that that's like the same uh things are happening again. But I think the main thing to to think about is like what is the outcome you're trying to drive? And at the time it was like opening up the number of outcomes and there's nothing more important at than the use of my time than to investigate as many outcomes as possible. Right? So if with that in mind, things like eating dinner were not as important was it was like dep prioritize because it was more important to talk to um either folks on uh other companies or investors or even like calming the employees uh because they were all in a very high stress state as well. And all I could think about was like how do I drive more outcomes here and increase the probability of success and again that sometimes means like dep prioritizing your health and your well-being. And you know obviously if you do that too much you don't want to overoptimize and you know pass away but so you want to think about what the balance is. So I think like yeah like if I were to look at it differently now it would be how do I not over compensate on the health side you know or over index on the de prioritization of health. I would definitely probably look back and be like probably should have had something to eat in between uh some of the uh conversations you know. — Cool. any other advice you have for like any other product that founders listening to this podcast like what would be maybe it's something where it's like hey this has been the most helpful thing that's helped us scale Windsurf or maybe it's this is how you could sell pretty quick to another company what would be your biggest piece of advice you'd share if this was like your two to three minute master class for any product head founder what would be that biggest piece of advice you'd give

### [32:32](https://www.youtube.com/watch?v=SY8d2oBXizE&t=1952s) Jeff’s Advice for Every Product Founder

— yeah you should always build something that is solving a very painful problem and you should always market it as such. I think a lot of mistakes uh some founders make is like I know how to use this technology and build a cool demo and make it like very flashy but in reality what you need is to drive revenue. You need a customer, an end user at the end of the day and you need to build something that is solving a problem specifically for them or at least they have to feel that way, right? And if you have to go to a very narrow focus, like there's only a few people that it solves a problem for, but 100% you're the only solution to do it, that is probably where you should you should start because agents are very general now and they can do a lot of things. Um, like Cloud Code and all these other products, they do a lot of things. If you build a product specifically for a narrow group of people that only will buy your product, that is a great place to start. And then once you drive that revenue, then you can start expanding and grabbing a larger share of the market. — Is there any like tips you'd have to get that like honed in other than just know your industry really well and see if like, hey, you interested in buying this? They're like, hell yeah, take my money. Or is there something else there? — You have to talk to a lot of potential prospects. You have to understand what are the painful things for them, right? I I'll tell you like with Kodium, we probably talked to maybe four or 500 different customers before we even like started closing a lot of deals in secession because that really made us understand the market and what to build. And even like downstream when you're doing um a pilot process and a sales process um there's a lot of things downstream that you you don't encounter until you have this repetition and you talk to a lot of customers such as pricing or support or deployment. like these things you really need to hone by just talking to a lot of the potential customers and I think that part people miss that that's probably something that people don't understand is a lot of work. — Yeah. And I think it's going to be a little bit more pronounced now too because the time to build something it's going down and it's a lot more fun to build something that you're kind of into versus oh I got to talk to another customer or potential customer that's probably going to say no. You have to think that's fun. You have to feel that it's fun to get a customer to say no to you and then it has to be like a game to try to flip that to a yes. If you don't think it's fun to get rejected, you should not be a founder. That is I think the perfect note to end on. Now, for people to find out more about what you're up to, where's the best places they can keep tabs or maybe even send you a message if they're like, "Hey, I've thought this really resonated for me and share some feedback. " Where would be the best places? Uh, I'm on Twitter at jeffwurf. Um, or you can catch me on LinkedIn. Either way, I'm uh those are the probably the places I post the most. — Okay, awesome. Well, thank you so much for coming on, Jeeoff. This has been a blast. Thank you for having me.

---
*Источник: https://ekstraktznaniy.ru/video/47531*