Learning Library

← Back to Library

Windsurf: AI Coding Startup’s Rapid Rise

Key Points

  • Founded in 2021 as Exofunction by MIT grads Varun Mohan and Douglas Chen, the company initially built GPU‑optimization tools before pivoting to AI coding assistance with Kodium.
  • After raising a $243 M Series C at a >$1.2 B valuation in 2024 (backers including Founders Fund and Kleiner Perkins), Kodium rebranded to Windsurf and launched an AI‑native development environment with the Cascade agent to compete with tools like Cursor, pricing its premium tier at $15 per seat versus Cursor’s $20.
  • In March 2025 Windsurf earned FedRAMP High certification, becoming the first AI coding tool cleared for U.S. government workloads and giving it a strategic edge in regulated industries.
  • By April 2025 the company reported roughly $100 M in ARR (up from $40 M months earlier), claimed over a million developers on the platform, and was notably stronger with enterprise decision‑makers than consumer engineers, positioning it as the #2 player behind Cursor.
  • The rapid rise of AI startups was underscored when Bloomberg broke news of OpenAI’s $3 B all‑cash acquisition in April 2025, illustrating the multi‑billion‑dollar M&A landscape that Windsurf now operates within.

Sections

Full Transcript

# Windsurf: AI Coding Startup’s Rapid Rise **Source:** [https://www.youtube.com/watch?v=qgQvci8NzM8](https://www.youtube.com/watch?v=qgQvci8NzM8) **Duration:** 00:21:24 ## Summary - Founded in 2021 as Exofunction by MIT grads Varun Mohan and Douglas Chen, the company initially built GPU‑optimization tools before pivoting to AI coding assistance with Kodium. - After raising a $243 M Series C at a >$1.2 B valuation in 2024 (backers including Founders Fund and Kleiner Perkins), Kodium rebranded to Windsurf and launched an AI‑native development environment with the Cascade agent to compete with tools like Cursor, pricing its premium tier at $15 per seat versus Cursor’s $20. - In March 2025 Windsurf earned FedRAMP High certification, becoming the first AI coding tool cleared for U.S. government workloads and giving it a strategic edge in regulated industries. - By April 2025 the company reported roughly $100 M in ARR (up from $40 M months earlier), claimed over a million developers on the platform, and was notably stronger with enterprise decision‑makers than consumer engineers, positioning it as the #2 player behind Cursor. - The rapid rise of AI startups was underscored when Bloomberg broke news of OpenAI’s $3 B all‑cash acquisition in April 2025, illustrating the multi‑billion‑dollar M&A landscape that Windsurf now operates within. ## Sections - [00:00:00](https://www.youtube.com/watch?v=qgQvci8NzM8&t=0s) **Windsurf: AI Dev Tool Evolution** - The transcript outlines the rapid rise of the startup Windsurf—originally founded as Exofunction and later rebranded from Kodium—a well‑funded AI coding assistance platform that pivoted to an AI‑native development environment, secured a $1.3 billion valuation, undercut competitors on price, and achieved FedRAMP High certification in early 2025. - [00:03:57](https://www.youtube.com/watch?v=qgQvci8NzM8&t=237s) **OpenAI Secures Windsurf Acquisition** - In early May 2025, after a letter of intent and exclusive negotiations, OpenAI finalized a $3 billion deal to fold Windsurf into its ChatGPT developer suite, coinciding with Anthropic’s withdrawal of Claude and mixed reactions from the developer community. - [00:07:27](https://www.youtube.com/watch?v=qgQvci8NzM8&t=447s) **OpenAI Deal Collapse Triggers Exodus** - The transcript outlines how the expiration of OpenAI's exclusivity agreement, compounded by Anthropic's strategic actions, forced users toward competitors and precipitated the collapse of a major enterprise deal. - [00:11:14](https://www.youtube.com/watch?v=qgQvci8NzM8&t=674s) **Google Deal Sparks Silicon Valley Trust Crisis** - The speaker argues that Google’s acquisition, driven by AI salary pressures, rewards founders for abandoning their startups and undermines long‑term trust in the Silicon Valley ecosystem. - [00:14:43](https://www.youtube.com/watch?v=qgQvci8NzM8&t=883s) **Cognition Labs Steps In** - The speaker explains how Cognition Labs is rescuing a Google‑sidelined developer platform (Devon/Windsurf), leveraging its FedRAMP certification for government contracts, reviving ties with Anthropic, and illustrating the extreme talent‑money dynamics driving Silicon Valley's talent wars. - [00:17:57](https://www.youtube.com/watch?v=qgQvci8NzM8&t=1077s) **Model vs. Tool Makers' Power Struggle** - The speaker outlines how model creators control roughly two‑thirds of the AI ecosystem’s power while tool makers retain about one‑third, using Windsurf’s acquisition and FedRAMP certification as examples, and predicts that code‑based agents will expand as a clear, rewardable use case. ## Full Transcript
0:00Wind surf is one of the most insane 0:02startup stories that I have run across 0:04in couple of decades hanging out around 0:07Silicon Valley and tech. We are going to 0:09talk about it and we're going to start 0:10by telling you the story of wind surf so 0:12you can get a sense of how long this 0:14saga has been going on. 2021 0:18company was founded as exofunction by 0:21MIT graduates Varun Mohan and Douglas 0:23Chen. The original focus was GPU 0:26optimization tools. So even then they 0:28were in the AI space. They then pivoted 0:30to AI coding assistance and launched 0:33Kodium. 2024 they raised series C. The 0:37valuation at the time was somewhere 0:38between 1.25 billion and 1.3 billion and 0:42the total raise to date is 243 million 0:45almost a quarter of a billion dollars. 0:47Investors at the time of the series C 0:49include Founders Fund and Kleiner 0:51Perkins. 0:52Late in 2024 after the series C, they 0:55had a major rebrand from Kodium to 0:57Windsurf and they launched an AI native 1:00development environment with the cascade 1:03agent intended to be a competitor to 1:05cursor. Everyone was seeing cursor take 1:08off. This was the point we were seeing 1:10bolt take off. Lovable was starting to 1:11take off. Windsurf was positioned in 1:14that ecosystem as a full stack app 1:17builder. It had a free tier and then it 1:20had a uh I think 15 buck per seat 1:23premium at the time versus cursors 20 1:25bucks. They're clearly undercutting on 1:27price. 2025 is where things get 1:30interesting. In Q1 on March 11th, 1:32Windsurf achieves Fed Ramp High 1:35certification, which if you don't know 1:36what it is, count yourself blessed. It's 1:39the first AI coding tool to be approved 1:42for US government workloads. That's what 1:44FedRamp High means. That gives it 1:47obviously a major leg up in regulated 1:50sectors and it plans to take that all 1:52the way to the bank. Early in April, 1:55Windsurf hits reportedly $100 million in 1:59ARR, far behind Curser, but still 2:01clearly in the number two position. It 2:03is the lift to the Uber played by 2:06Curser. This is up from $40 million just 2:09a few months before. and they claim a 2:12million plus developers using the 2:13platform with 350 plus enterprise 2:16customers. One of the things I will call 2:18out here anecdotally is I hear wind surf 2:21more with enterprise customers than I 2:24hear it with individual engineers. And I 2:26think that's a very interesting uh flip 2:29from cursor because cursor I hear from 2:31engineers all the time and I hear it as 2:34yes they're going to consider it but 2:36it's not that surprising if you consider 2:38the ubiquitous number one. Windsurf has 2:41better name recognition with decision 2:43makers and punches above its weight in 2:45decision-making circles. And I think 2:46that's really interesting and that will 2:48continue to play out as we as we 2:49continue the story here. April 16th, 2:522025, Bloomberg breaks the OpenAI 2:55acquisition news. The reported price $3 2:58billion allcash deal. It would be 3:01OpenAI's largest acquisition ever. We 3:04are at a stage in AI and private 3:06companies where you can have 3:07multi-billion dollar acquisitions by a 3:10startup of another startup. That is how 3:12big AI is right now. Obviously, if 3:15you're at Windsurf, you begin doing the 3:17math on your equity. You want to say 3:19like, how much do I get? What is my vest 3:22date? Everyone goes and looks at their 3:23RSUs. I've done this sort of song and 3:25dance before. They do the math. They try 3:27and come up with something. I will tell 3:28you as someone who has lived through 3:29these things a couple of times. Do not 3:32do the math yourself. you will end up 3:34with a number that is by definition 3:36incorrect because you will forget the 3:38taxes or you will forget the fees or you 3:40will forget the like delay in the payout 3:42due to some sort of uh clause that they 3:46write in the contract. There are all 3:47kinds of ways to change the math that 3:49are legal. It is just there's no point 3:51doing the math yourself. Just don't 3:53bother. 3:55Early May, the letter of intent is 3:57signed and the exclusive negotiation 4:00window begins. 4:02internal memos at this point leak 4:05showing that there's a plan to fold 4:07Windsurf into the chat GPT dev suite. 4:10This is also when Anthropic 4:13is going to start withdrawing Claude 4:16from Windsurf causing a lot of bad 4:19feelings from the developers using 4:21Windsurf. I was a user of Windsurf at 4:23the time. I was very unhappy that Claude 4:25left. But I get it. If you have 4:27something called chat GPT dev suite 4:29coming, the last thing you want is to 4:32let Claude's output tokens and Claude's 4:34data into that environment because 4:36really what you're doing when you're 4:37coding is you have an AI writing code 4:40and you have an engineer rewarding that 4:43one way or another. Either telling the 4:45code output, hey model, great job on the 4:47code output, good job, thumbs up, and 4:50then you have a reward signal or hey, 4:52this didn't work. Can you try it this 4:54way? And then you have a correction 4:55signal. Both are highly valuable signals 4:57if you are in the AI arms race and 4:59Enthropic did not want those signals 5:01going to chat GPT. So OpenAI reportedly 5:06agreed to definitive terms with Windsurf 5:08around this Chat GPT dev suite idea. 5:12May 6th, 2025, a definitive agreement is 5:15announced which is a step beyond just 5:17the negotiation window and the letter of 5:19intent. A step beyond the uh reported 5:22price negotiated initially. the $3 5:24billion price was confirmed again. Uh 5:27this is a rare case where the price 5:29didn't change and Windsurf employees 5:31rightly are celebrating at this point. 5:33They think they've got it made, but we 5:34aren't done yet. Early June negotiations 5:38hit a major snag. Microsoft raises IP 5:42concerns under 2023's agreement that 5:45OpenAI signed with Microsoft back when 5:47OpenAI had many fewer cards than it 5:49holds today. Microsoft has rights to all 5:53OpenAI tech, including acquisitions, and 5:56it looks like Satcha Nadella played his 5:58hand. The rumor is that OpenAI did not 6:02want to give the IP for Windsurf to 6:05Microsoft. There has been, I don't want 6:07to call it a divorce, that sounds 6:09dramatic. There has been a soft 6:11separation between Microsoft and Open 6:13the Eye. Let's use that term. Very 6:15polite. We're all adults here. And there 6:18has been a little bit of tension and 6:20this is an example of that tension. 6:21Basically, OpenAI has become more 6:23powerful in the relationship since the 6:25agreement was signed. They're flexing 6:27their wings. In this case, they flexed 6:28their wings and said, "You know what? 6:30We're acquiring Windsurf and no 6:31Microsoft. We don't want you to have the 6:33IP." And Satcha basically said, "Talk to 6:35my Microsoft lawyers." Which if you've 6:37ever talked with Microsoft lawyers, you 6:39don't want to do. So, at the end of the 6:42day, Satcha was right. Microsoft had 6:45legal rights. GitHub Copilot directly 6:47competes with Windsorf and he was right 6:49to worry about it and Microsoft refused 6:51to grant an exception to their IP 6:54rights. 6:56June three and Anthropic formally 6:59restricts Windsurf's claw access. This 7:02is during the OpenAI exclusivity period 7:04and they are probably doing it to sort 7:06of see if they can scotch the deal in 7:08addition to uh in addition to the token 7:10impact that I discussed earlier. Um, 7:15as Jared Kaplan put it, it would be odd 7:18for us to sell Claude to OpenAI. That's 7:20that's exactly sort of the frame that I 7:21made, but it's it's much more sus 7:23succinct. 7:25So, this forces winfur users to migrate 7:27to competitors. Now, you have a deal 7:28that isn't done yet, and you have users 7:30leaving the platform. Enthropic is 7:32quietly happy uh because OpenAI is kind 7:35of getting screwed on this deal. Mid to 7:38late June, things begin to unravel. 7:40OpenAI desperately attempts to 7:42restructure the deal. The wind surfboard 7:44begins to get worried about losing 7:45enterprise and government contracts as 7:47their core product starts to collapse 7:49because Claude is starting to pull out. 7:51Claude is, if you don't know, one of the 7:53best coding agents out there. The deal 7:55is effectively stalled. It's not 7:58publicly dead. And there's a difference 8:00there. 8:02July 11th, that was this Friday, 7 a.m. 8:06Pacific, the OpenAI exclusivity period 8:10officially expires. No renewal is 8:12announced and the deal is confirmed to 8:14be under review, which is code for in 8:17trouble. 8:18By the morning, news breaks that the 8:20OpenAI deal has collapsed. Microsoft's 8:23IP issues are cited as the primary 8:26cause. Although if you look into it and 8:28you look at Anthropic's power move, I 8:30would argue that's also a contributing 8:32factor. If you start to lose the most 8:34powerful coding agent, it's hard to make 8:35the case that you're a very powerful and 8:37$3 billion worth development platform. 8:39This underscores the tension between 8:41model makers and the tool layer right 8:43above them. The information, which is a 8:46tech publication, now reports talks have 8:48ended on tech sharing concerns. By 300 8:52p.m. that day, Friday, July 11th, Google 8:54has announced a $2.4 billion deal. They 8:58got on the phone right away. 9:00They structure a reverse aqua hire with 9:03a licensing deal. Now there I have 9:05wondered whether this is effectively a 9:08way to get around FTC examination of the 9:12deal because it's not technically an 9:13acquisition. It is a very unusual 9:16acquisition format that isn't an 9:19acquisition because it's an aqua hire. 9:21They do not acquire the company. Instead 9:25they hire the CEO Varun Mohan and the 9:27co-founder Douglas Chen and about 40 of 9:29the best engineers R&D employees. They 9:32all get paid bank. They get paid. They 9:35split between them 2.4 billion just for 9:38the talent and a non-exclusive license 9:41to wind surfs tech. Non-exclusive. You 9:44understand? Mostly that is 2.4 billion 9:47for 42 people. 9:50That's a good deal. Like divide out 9:51those numbers. That is a good deal. The 9:54remaining quarter of a thousand 9:55employees, the remaining 250 employees 9:57are not included in this deal. Don't 9:59know about this deal. Aren't 10:00participating in this deal. their equity 10:03is more or less worthless because what 10:05they get is $100 million in remaining 10:08cash in the company uh and they don't 10:11have the option to compete on their tech 10:14exclusively because Google can now 10:15license their tech and they don't have 10:18their best R&D employees. They don't 10:20have their founders and they are left 10:23screwed for lack of a better term. Now I 10:25will say not everyone views it that way. 10:27I thought that founding engineer well 10:29founding engineer engineering uh maven 10:32Gurgly Oros had a really good point when 10:36he said honestly as bad as it may be if 10:39you're in it the outcome for wind surf 10:42across the board is better than most 10:44startup outcomes and he is right most 10:47startup outcomes 90% plus go to the wall 10:50they failed the startup equity is worth 10:52zero the fact that you got paid 10:56and that some of your employees and your 10:58founders got something is still better 11:02than the zero outcome you would expect 11:03on most startups. That may be true, but 11:06I think what Girkley misses is the 11:09trustbreaking nature of this choice. 11:11There has been an implicit contract in 11:14the valley where founders are in it for 11:17the long haul with their companies. If 11:20that starts to break down because the AI 11:23salary pressures are so insane that it 11:26is better to be an employee than to be a 11:28founder at a major R&D firm like Google 11:32for AI. 11:34We have cracks in the Silicon Valley 11:37ecosystem. 11:39It does not sit well with me that a 11:42founder would be rewarded for walking 11:45away from their startup and leaving 11:47their employees in the lurch. So as much 11:48as I agree with Gurgaly that on paper 11:51this is better than the outcome the 11:53company could have had I don't think it 11:58is a good thing that builds trust in the 12:00ecosystem long term. I think it's 12:02negative for the ecosystem. I don't 12:03think Google did a service to the valley 12:05here. And I get maybe why they felt they 12:08had to do it because Google unlike 12:09smaller startups like OpenAI would not 12:11be under FTC scrutiny the way Google is. 12:13And so Google had to tread very 12:15carefully. And that may be part of why 12:16the Aqua Hire deal happened. 12:19But we aren't done yet. 12:22By the evening, 12:24Logan at Google, Logan Kilpatrick, 12:26tweets a welcome and everything is 12:28official. Controversy starts to explode. 12:31Details of the structure leak on July 12:3312th, Saturday. The 2.4 billion leaks. 12:37Vested employees got cash payouts. 12:39Employees with less than 12 months 12:41tenure get nothing, which is actually 12:42very typical in startups. And the 12:45remaining company has no core team, 12:46which I discussed. In the afternoon on 12:48Saturday, it's announced that Jeff Wang 12:50will be the interim CEO. The company 12:52states their SAS focus will continue and 12:55emphasizes the Google license is not 12:57exclusive. 12:59Social media explodes. 13:02Uh, lot of commentary on employees 13:05counting out sort of their rewards they 13:07were going to get and then getting 13:08nothing. Uh, channel partners begin to 13:10pull support for for Windsurf. # 13:13Windinssurf betrayal starts to trend. By 13:16July 13th, Sunday, more internal details 13:19leak. Apparently, there was a rambling 13:21CEO Zoom speech announcing the deal. 13:23Employee anger begins to peak. 13:26By the afternoon, industry analysts are 13:28weighing in. Predictions of bankruptcy 13:30within 12 months. Remaining company 13:33called a zombie. Enterprise customers 13:35advised to migrate. This is all very 13:37bad. If you are still at Windsurf, 13:38right, for the 250 employees, this has 13:40been a weekend from hell. And I I cannot 13:43emphasize this enough. I would not want 13:45to be you guys. I I am hoping that you 13:48are in a better spot now. And we're 13:49going to get to why. July 14th, Monday, 13:52Cognition Labs, the makers of Devon, 13:55announced an acquisition of the 13:57remaining parts of Windsurf. This 13:58includes their IP, their product, their 14:00trademark, their brand, their training 14:01data, and all remaining 250 employees. 14:05The key difference here is there is 100% 14:08employee financial participation, full 14:10vesting, acceleration, and wave cliffs. 14:11In other words, Cognition Labs saw what 14:14Google did and said, "We have an 14:15opportunity to step in as the good guy 14:17here. We can be the good guy. We will 14:20acquire Wind Surf. We'll bring everybody 14:23aboard. It doesn't matter whether you 14:24vested or not. If you've worked less 14:26than a year or not, which by the way, 14:27when you're in a startup and it's hiring 14:29like on a power law like this, most of 14:31your employees by definition are less 14:33than a year in because you're hiring so 14:36fast. So, I would expect for most of 14:38these employees, waving the cliff was a 14:40big deal. 14:42Essentially, Cognition Labs makes them 14:43whole where Google screwed them over. 14:46And it's a nice move for them because 14:47they're basically saying we have an 14:48agentic developer in Devon. Now we have 14:50tools for a development environment in 14:52Windsurf. In theory, if they can get 14:54themselves sorted out, they still have 14:55that Fed ramp high certification which 14:57could get Devon into government work. 15:00There's some upside here for Cognition. 15:02It wasn't just that they were being 15:03charitable. Now, of course, the pricing 15:04isn't disclosed. I can't imagine it was 15:06$3 billion, not least because Cognition 15:09doesn't have the money that OpenAI had. 15:12By afternoon, 15:14more conflicting reports start to 15:15surface. 15:17Uh, Windinsurf reposts the Cognition 15:20announcement. The show is going on. 15:22Cognition emphasizes that we're friends 15:24with Anthropic again, meaning that they 15:26think they're going to be able to get 15:27Claude back into Windsurf to revive it. 15:31We will have to see how this all goes. 15:33At the end of the day, 15:36I want to draw a couple of lessons. One, 15:39this is underscoring the 15:43skewed nature of the talent wars and the 15:45power law for talent wars in Silicon 15:46Valley in a way I have never seen 15:48before. If you can pay a founder who is 15:51already a multi-millionaire so much 15:53money that founder finds it worthwhile 15:55to abandon their fiduciary 15:57responsibilities to their company, 15:59potentially expose themselves to a 16:01lawsuit because there there's some board 16:04level responsibility here to uh work at 16:06the best interest of the company and 16:08it's arguably not the best interest of 16:10the company to walk away from the 16:11company and join Google for millions of 16:13dollars. So they're exposing themselves 16:15to that and on top of that they are 16:20frankly breaking a code of honor in the 16:23valley that has been true for decades. 16:25That's still worth it to them because 16:26Google's paying them so much. And that 16:28underscores how massive these pay 16:30packages are. Over like a weekend ago, 16:33Zuck got the I think the head of AI at 16:37Apple. I don't know why you would go to 16:38Apple for that, but he got the head of 16:40AI at Apple for $200 million, which is 16:42more than Tim Cook at Apple makes, I 16:44think, depending on how you count the 16:46equity investing. Anyway, it's a ton of 16:48money. 16:50When you make that much, you break the 16:52usual incentives at Valley. If you can 16:54be a $200 million employee, why would 16:56you ever found? 16:58And I think that we are starting to see 17:00some of that dislocation. The second 17:01thing that I want to call out is that 17:04there is massive dislocation in the tool 17:08layer and that we should view coding 17:11agents, the coding environment as the as 17:14the tip of the iceberg for the 17:16disruption that model makers can pose. 17:19Because at the end of the day, if you 17:22are a model maker and you believe good 17:25models will get better over time, you 17:27should be able to eat more and more of 17:29workflows over time just by delivering 17:31good models. And you can do something 17:32like, I don't know, clawed code, which 17:34is frankly just a terminal, and compete 17:37with cursor, 17:38you're going to eat some of these tools 17:40for lunch over time, and you kind of 17:42don't care. And so we are seeing this 17:44tectonic struggle between the toolmakers 17:47that are trying to build on top of this 17:48modelmaker tech and the model makers who 17:51are really incentivized to grab as much 17:53territory as they can. 17:56Modelm makers hold the cards here as you 17:57saw with Claude cutting off access to 18:00Windsurf but they don't hold all the 18:02cards. And so one of the things that's 18:04super interesting about this dynamic is 18:06that it's sort of a twothirds one-third 18:08situation where you might say that 18:10twothirds of the power sits with the 18:11model makers, but the tool layer has a 18:13third of the power. At the end of the 18:15day, Windsurf still got acquired. 18:17They're still going to be a viable 18:18company. The Fed Ramp High certification 18:20did not go to Claude Code. It went to 18:22Windsurf. And so tool makers are still 18:25going to get their day in the sun. 18:27They're still going to get their 18:28dollars, but the way they negotiate the 18:30relationship with model makers is going 18:32to continue to be fraught because the 18:33model makers have a lot of incentives to 18:35make sure that their models continue to 18:37get better and the toolmakers are going 18:39to have to negotiate those incentives 18:41while still serving customers. And 18:43that's sort of the story that tension is 18:45the story of Windsurf over the last 18:48couple months. The third thing I want to 18:50call out is that we are going to see 18:51more and more agents in code. And again, 18:53that is a cuttingedge tip of the spear 18:56for where we will see them other places. 18:58The reason code is leading the way here 18:59is because code is easy to reward. Code 19:03runs or it doesn't. It's easy for model 19:05makers to figure out whether their code 19:06worked or it didn't. And that's true for 19:08development environments. It's true if 19:10you launch your own agent. Cursor 19:11recently launched their own model that 19:13codes. Uh and it's true if you have an 19:16agent that runs on code. All of the 19:18stuff we're seeing in the development 19:19environment now, this whole conversation 19:21around wind surf is an early indicator 19:25of where the rest of the industry is 19:26going because it's an easier problem for 19:28AI to solve. 19:31When you think about the next 12 to 15 19:33to 18 months, 19:35Devon acquiring windsurf means more work 19:38on agents in code. It means the agentic 19:41agenda for lack of a better term. Not 19:42that it's a secret cabal, but the 19:44agentic goal of like getting AI agents 19:46with tools and policy and guidance to 19:49actually do important coding work that 19:51is moving forward. Devon is, I would 19:53expect, going to be fully integrated 19:55into Windinsurf and we're going to see 19:57more of this hybridized uh motion where 20:00you have both a development environment 20:01and an agent management environment for 20:03engineers in the same place. Cursor 20:05heavily going that way as well. That is 20:08going to lead to new surfaces, new 20:10experiences. The mobile environment has 20:12become a much bigger deal for developers 20:14than it ever was. It used to be you 20:16could just build for the desktop because 20:17always developers would be on the 20:19desktop. Not anymore. If you're running 20:20agents, you can be on your mobile. 20:22There's a lot of really weird 20:24implications and that's very typical of 20:26what happens when you have an underlying 20:28technology shift like we are all living 20:30through right now. Fundamentally, the 20:32story of Windsurf is the story of the 20:34evolution of LLMs from being just sort 20:38of autocompletes for coders to being 20:40genuine coding agents to being genuine 20:42uh assistants in code. And what happens 20:46when valuations start to shake loose at 20:48that point? That's why OpenAI went for 20:51Windsurf in the first place. That's why 20:53they got the Fed Ramp High 20:54certification. That's why Devon went for 20:56them. And that is also why that tech is 20:58why Google needed them because Google 21:00doesn't have a development environment 21:02that is popular with developers and if 21:04they're going to get Gemini to win long 21:06term they need it. So that's the story 21:09of Windsurf. I am sure there will be 21:12more to the story. It's been one of the 21:13most eventful of the last couple of 21:16decades I would say and I'm curious what 21:19do you think you would do in Windsurf 21:21shoes?