Jul 28, 2025 · 3h 53m · latent-space
🕰️ The Oral History of Windsurf (ft. Varun Mohan, Scott Wu, Jeff Wang, Kevin Hou, Anshul R)
Varun Mohan · 1h 9m spoken Kevin Hou · 45m spoken Scott Wu · 41m spoken Anshul Ramachandran · 18m spoken Shawn Wang · 17m spoken Alessio Fanelli · 9m spoken Jeff Wang · 3m spoken
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
This comprehensive retrospective presents the oral history of Windsurf and Cognition AI, charting the evolution of AI developer tooling from low-level GPU infrastructure and code autocomplete to autonomous agents and their landmark merger.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 12.8% of the talking time here. How this is scored →
The hosts as informed peer 3.2 Guest teaching 3.3 Guest disagreement 1.1 The hosts pushing back 1.0
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement â–¶ 1:50:50 Anshul rejects standard SWE benchmarks
Anshul emphatically dismisses industry-standard benchmarks like SWE-bench and HumanEval as bogus, arguing that optimizing against them leads to suboptimal real-world architectures.
Hardest push from the hosts â–¶ 3:51:05 Swyx addresses the controversial acquisition mechanicsSwyx directly confronts Scott Wu and Jeff Wang with delicate reported details regarding Google's $2.4B deal, Cognition's $300M buyout, and broader Silicon Valley anxiety over broken founder-employee contracts.
Biggest teaching moment â–¶ 28:20 Varun breaks down Chinchilla vs serving realityVarun educates Swyx on why theoretical Chinchilla compute-optimal ratios fail in production, explaining how hardware memory and serving latency dictate heavily over-training smaller models.
The host holds their own â–¶ 2:02:25 Swyx identifies ColBERT late interaction patternSwyx demonstrates deep academic and technical grasp by connecting Codeium's proprietary real-time distributed retrieval system directly to Omar Khattab's ColBERT late interaction research.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Introducing Varun Mohan and ExaFunction's Background | 6 | 5 | 1 | 1 | Alessio and Swyx demonstrate strong technical literacy around deep learning infrastructure, referencing model flop utilization (MFU) and dynamic multiplexing. Varun Mohan explains the low-level GPU utilization challenges and operator fusion techniques that drove ExaFunction's inception. | |
| Shifting Focus from Infrastructure to Codeium | 5 | 4 | 2 | 1 | Varun describes pivoting from pure infrastructure to building Codeium after seeing Copilot handle complex C++ variadic templates. The hosts probe on data sourcing (scraping vs The Pile) and the strategic rationale behind offering free code completion. | |
| Trade-offs in LLM Product Design: Latency, Quality, Correctability | 6 | 8 | 3 | 2 | Varun educates Swyx on the nuances of Chinchilla compute-optimal training, explaining that serving constraints and memory limits force production models to be heavily over-trained. He introduces the critical product framework of latency, quality, and correctability. | |
| Emergence, Context Scaling, and Data Curation | 5 | 7 | 3 | 1 | Varun outlines advanced retrieval architectures like DeepMind's RETRO, arguing that existing vector database solutions are cost-prohibitive and unsuited for multi-terabyte LLM scale. Swyx playfully points out Varun comparing Codeium's in-house approach to early Google. | |
| Production A/B Testing and Domain-Specific Specialization | 5 | 6 | 1 | 1 | Varun discusses hardware networking realities, including NVLink bandwidth and the massive optimizer state required for synchronous distributed training. The hosts ask about the EleutherAI community and future expectations for domain-specific models. | |
| Cognition AI Unveils Devin, the First AI Software Engineer | 0 | 0 | 0 | 0 | Keynote and launch audio presentation featuring Scott Wu introducing Devin and Kevin Hou introducing context retrieval challenges. Host scores are 0 as this is recorded presentation audio. | |
| Introducing mQuery and Codeium's Vertical Integration | 0 | 0 | 0 | 0 | Conference presentation by Kevin Hou detailing Codeium's mQuery system, which runs thousands of parallel LLMs over codebases rather than low-dimensional embeddings. Host scores are 0 for monologue presentation. | |
| Autonomous Vehicle Analogy and Shipping Practical AI | 0 | 0 | 0 | 0 | Kevin Hou presents an analogy between autonomous vehicles and AI coding tools before transitioning to Scott Wu's live demo at the AI Engineer World's Fair. Monologue presentation format. | |
| From Text Completion to Autonomous Agent Workflows | 0 | 0 | 0 | 0 | Scott Wu's conference presentation on agentic workflows versus text completion, highlighting multi-file edits, debug iteration loops, and async collaboration via Slack. Monologue presentation format. | |
| Scott Wu Audience Q&A on Access, Sandboxes, and Interns | 0 | 0 | 0 | 0 | Audience Q&A with Scott Wu followed by the official launch audio for the Windsurf editor featuring Varun Mohan and Anshul Ramachandran. Host scores remain 0 for conference recording. | |
| Latent Space Podcast in the New Codeium Office | 5 | 5 | 1 | 1 | Swyx and Alessio record in Codeium's new Mountain View office. Varun highlights enterprise reality where most Fortune 500 companies use GitLab/Bitbucket and JetBrains rather than GitHub and VS Code. | |
| Overcoming VS Code API Limitations and Defining Cascade | 5 | 7 | 4 | 2 | Varun and Anshul explain why they forked VS Code to build Cascade, having previously had to dynamically generate PNGs to bypass UI limitations. Anshul forcefully dismisses conventional benchmarks like SWE-bench as bogus compared to production masking evaluations. | |
| Command Execution Safety, Sandboxing, and State Rollbacks | 7 | 5 | 2 | 2 | The conversation explores sandboxing, trajectory checkpoints, and execution safety. Swyx demonstrates sharp expertise by drawing a direct technical parallel between Codeium's retrieval architecture and Omar Khattab's ColBERT late interaction method. | |
| Pricing Philosophy and the Enterprise Value Distribution | 5 | 5 | 1 | 2 | Varun explains why individual developer subscriptions ($10/mo) are primarily cost-covering while the real enterprise margin lies in high-retention Fortune 500 contracts with heavy compliance needs. | |
| Community Feedback and the Enterprise Windows Reality | 5 | 4 | 1 | 1 | Alessio reviews Anshul's 2022 blog post, grading the predictions on context scaling, UX, and first-party models. Varun discusses building for the 80%+ enterprise developers on Windows and WSL. | |
| Grounding AI Development in Realistic Feedback and IDE Data | 6 | 7 | 4 | 2 | Varun passionately argues that having internal AI skeptics with autonomous vehicle backgrounds keeps engineering grounded against superficial Twitter hype. He delivers a strong defense of building proprietary core competencies (like custom inference runtimes) rather than relying on vLLM. | |
| Scaling Technical Go-to-Market and Forward Deployed Engineering | 5 | 6 | 1 | 1 | Varun and Anshul explain the challenge of scaling enterprise technical sales and deploying Forward Deployed Engineers (FDEs) who deeply understand RAG and custom infrastructure. | |
| Windsurf Principle 1: Trajectories and Shared Timelines | 0 | 0 | 0 | 0 | Keynote presentation by Kevin Hou in New York breaking down Windsurf's core architectural principle of unified human-AI trajectories. Monologue presentation format. | |
| Windsurf Principle 3: Scaling with Intelligence and Sunsetting Chat | 0 | 0 | 0 | 0 | Kevin Hou discusses sunsetting chat and rules files in favor of inferred context, followed by Scott Wu's talk on Moore's Law for AI agents (doubling capability every 70 days in code). Monologue presentation format. | |
| Advancing Devin: Codebase Understanding, DeepWiki, and Autonomous Backlogs | 0 | 0 | 0 | 0 | Scott Wu's talk tracing Devin's progression from repetitive migrations to DeepWiki codebase understanding and automated backlog clearing. Monologue presentation format. | |
| Keynote: The 10 Waves of Windsurf and Full-Lifecycle Integration | 0 | 0 | 0 | 0 | Kevin Hou's keynote on the 10 waves of Windsurf, detailing integration across Google Docs, Figma, Jira, MCP tools, and Netlify deployments. Monologue presentation format. | |
| Always-On AI and Introducing SWE-1: A Specialized Engineering Model | 0 | 0 | 0 | 0 | Kevin Hou introduces SWE-1, a specialized software engineering model trained on real interactive dev trajectories rather than purely static code generation benchmarks. Monologue presentation format. | |
| Behind the 72-Hour Cognition and Windsurf Merger | 5 | 4 | 1 | 1 | Exclusive interview with Scott Wu and Jeff Wang detailing the high-speed 72-hour merger between Cognition and Windsurf following Google's licensing deal. Both founders recount working sleeplessly with global legal teams over a single weekend. | |
| Product Roadmap Convergence: Combining High-Touch IDE and Async Agents | 7 | 5 | 2 | 6 | Swyx directly raises the sensitive industry rumors regarding Google's $2.4B licensing acquisition, the $300M Cognition deal, and concerns over the founder-team covenant in Silicon Valley. Scott Wu and Jeff Wang address the IP boundaries, data exclusivity, and team equity protections. |
Statements from this episode (104)
Mohan: GPU container sharing limitations leave hardware heavily idle
“For most people, one of the things about CPUs that's really nice is with containers, right? You can end up having a single node and you can place many containers on them and all the containers will slowly start eating the compute. It's not really the same with…”
Fanelli: ExaFunction customer cut compute costs 97% on single GPU
“And I saw one of your customers, they went from. 30 clients to just one single GPU and they cut costs by 97%.”
Mohan: Enterprises should fine-tune off-the-shelf models over custom architectures
“For a vast majority of enterprises, they should probably be using something off the shelf, fine tuning BERT models. If it's a vision, they should be fine tuning resonant or using something like clip, like the less work they can do the better.”
Mohan: Few products match Copilot due to compute-intensive low latency
“There's like a gap there where the number of products that have hit that co-pilot spot is actually not that high. A lot of them are kind of like wait and, you know, just generate a lot of stuff and see what happens because one is clearly more compute intensive…”
Swix: Copilot report estimates 60-70% of AI-generated code is checked in
“There's a report this morning from Copilot where they were estimating the key tabs on amount of code generated by a Copilot that is then left in code repos and checked in. And it's something like 60 to 70%.”
Mohan: AI coding tools create a self-fulfilling loop by changing developer behavior
“Once you start using products like this, where in the beginning there's like skepticism, like how, how valuable can it be? And suddenly now like user behavior fundamentally changes so that now when I need to write a function, I'm like documenting my code more …”
Mohan: Codeium scraped public code because The Pile was too small
“We ended up basically using a lot of open, I guess, permissively licensed code in the public internet, mainly because I think also the pile is, is fairly a small subset. I think maybe after we started, there was the stock that was also came to be, but for us, …”
Mohan: Off-the-shelf serving frameworks leave significant FLOP utilization on the table
“The open source serving. Offerings are just, I will say not great in that they aren't customized to transformers and these kinds of workloads where I have high latency and I want to like batch requests and I want to batch requests while keeping latency low. Bu…”
Mohan: Codeium bootstrapped with open-source models before training custom FIM models
“Early on, we were able to leverage open source models, sort of bootstrap it internally within our company. But then to ship, we finally had some requirements like, Hey, we want this model to have fill in the middle capabilities and a bunch of other things. And…”
Mohan: Codeium reached 10k users and 5% daily growth in late 2022
“So right now we have, we probably have over 10,000 users and thousands of daily actives and people come back day over day. Our growth is like around, you know, four to five percent day over day right now.”
Mohan: Codeium quality matches Copilot and drives user churn
“The product is actually one of those products where even use Copilot and use us, it's hard to tell the difference actually. And a lot of our users have actually churned off of Copilot.”
Mohan: AI startups should bootstrap on existing APIs unless fine-tuning on large datasets
“Unless you are, you probably should always bootstrap on top of an existing API, right? Because like, even if you were to, the only reason why we didn't is because we knew that this product was actually buildable. Probably if we worked hard enough to train a mo…”
Mohan: LLM products depend on latency, quality, and correctability
“And if I were to go out and think about like the three categories of like an LLM product, it's probably like latency, quality, and correctability. The reality is, you know, if I were to take a product like Copilot or Kodia, the latency is very low. The quality…”
Mohan: Automated PR generation will require specialized models trained on diffs
“A lot of things people are excited about right now are I write a comment and it generates a PR for me. And that's like really awesome in theory. I think that's like a really cool thing. And I'm sure at some point we will be able to get there. That will probabl…”
Mohan: Codeium uses heuristics to detect developer exploration versus acceleration modes
“And we have heuristics in terms of capturing
Whether or not you're in acceleration versus exploration mode.”
Mohan: Generating subsequent code blocks increases cognitive load during acceleration mode
“And a good heuristic is let's say you're inside like a basic block of a piece of code. Let's say you're inside a block of code or an if statement. You're probably already in acceleration mode and you would feel really bad if I started generating the else claus…”
Mohan: Harvey proved value by signing a 3,500-lawyer enterprise contract
“It's clear that they've demonstrated some amount of value because they've been able to close a fairly sizable enterprise contract. It was like a firm with 3500 lawyers, something nuts, honestly, very cool.”
Mohan: Codeium will not expand outside code into legal AI anytime soon
“So it's probably going to be a while before we even consider leaving code. Like that's going to be a big step because the amount of learning we need to do is going to be high. We need to get users, right? We've learned so much from our users already. So yeah, …”
Mohan: Codeium trains smaller, compute-suboptimal models for lower inference latency
“And for our case, our models that we train internally, they might not be the most compute optimal. In other words, we probably could have had a better model by making it larger, but the trade-off would have been latency. We know what the impact of having highe…”
Mohan: Marginal loss improvements unlock massive gains in model reasoning
“Like, small wins at the margins are massive wins in terms of IQ. Like, it's harder to get those, and they don't look as big, but they're, like, massive wins in terms of reasoning. They can now do chain of thought, all these other things.”
Mohan: Current vector databases are unviable for terabyte-scale retrieval
“I'm actually not convinced that the current breed of embedding database companies are like ready for what the future sort of looks like, because I think I'm just looking at their pricing, how much it costs per gigabyte, and it's prohibitive at the scale we're …”
Mohan: Codeium to build embedding database infrastructure in-house
“I will just say for us, we will probably just build this in house to start with. And that's because I think the technology probably isn't there yet. And I think if the technology isn't there yet, like waiting on point solutions to come up is a lot harder than …”
Mohan: HumanEval benchmark scores are inflated due to GitHub training contamination
“One of the issues that ends up coming up with things like human eval is contamination, because a lot of these things that train models end up training on all of GitHub. GitHub itself has human eval. So they end up Training on that, and then the numbers are arb…”
Mohan: Bearish on outsourced eval startups because AI companies must own evaluations
“And I guess maybe one of the things I'm a little bearish on is If another company comes out and solves eval properly for a bunch of different verticals, what was the company that they were selling to really doing? What are they really doing at that point? If t…”
Mohan: Evaluating coding models requires hundreds of thousands of completions
“It's honestly really hard to get signal unless you have like a non-trivial amount of usage. Non-trivial meaning you're getting, you're doing hundreds of thousands of completions, if not millions of completions.”
Mohan: Training optimizer state requires 14x model parameter size in memory
“The optimizer state. So while you're training is 14 times the size of the model. So in this case, if it's like one hundred seventy billion parameters, it's probably, I'm not great at mental math here, but that's probably around 2.5 terabytes to just store the …”
Hou: Codeium surpassed 1.5M downloads across 40 IDEs
“We've been downloaded over a million and a half times. We're one of the top rated extensions across the different marketplaces. And to reiterate, reiterate, we offer free unlimited autocomplete chat and search across 70 different languages and 40 different IDE…”
Hou: Codeium is highest-rated dev tool in Stack Overflow survey
“We are the highest rated developer tool as voted in by developers in the most recent stack overflow survey. And you'll note that this is even higher than tools like chat GPT and GitHub copilot.”
Kevin Hou: Traditional vector embeddings have hit a performance ceiling
“It's evident through the years that we're actually hitting a ceiling on what is possible using these traditional vector embeddings. And over time, even the biggest models are approximating to around the same level of performance.”
Kevin Hou: Code search requires multiple needles, unlike standard benchmarks
“Benchmarks like the one that I showed you before heavily skew towards this idea of needle in a haystack. It's the idea that you can sift through a corpus of text and find some instance of something that is relevant to you. Note, it is only one single needle. S…”
Hou: Public retrieval models underperform on git commit benchmarks
“We threw some of the currently publicly available models at this notion of retrieval, this idea of using commit messages, and we found that there is reduced performance. They're unable to reason over specifically code, but then also specifically this kind of r…”
Hou: Codeium inference costs 1/100th of competitors by avoiding third-party APIs
“It's that idea that our computation is one 100th of the cost of the competitors. We are not using APIs, and as a result, our customers and our users actually get 100 X the amount of compute that they would on another product.”
Hou: Codeium's mQuery runs thousands of LLMs in parallel for millisecond latency
“MQuery runs thousands of LLMs in parallel, so the user can start streaming in code within seconds, not minutes, not hours, Seconds and oftentimes milliseconds.”
Hou: Future AI tools will generate docs, reviews, and UIs from Figma
“But in the future, they're going to generate documentation. They're going to generate commit messages, code reviews, code scanning. They're going to take, you know, Figma artboards and convert them into component into UIs that were built by your own components…”
Hou: Embedding-based retrieval is just a temporary heuristic for limited compute
“And so at Codium, we believe that this embedding based retrieval is the heuristic. We should be planning for AI first products, throwing large models at these problems so that AI is a first class citizen.”
Wu: Devin built many of Cognition's API integrations and internal dashboards
“A lot of the API integrations that Devin has were built by Devin. You know, a lot of our own internal dashboards and metrics tracking within Devin were actually also built by Devin.”
Scott Wu: AI is shifting from text completion tools to autonomous agents
“The first wave of generative AI is what I generally call these text completion products, right? And, you know, that makes a lot of natural sense if you think about it that obviously the interface of a language model is text completion, right? You give it a pre…”
Scott Wu: Software engineering is investigative work suited to autonomous agents
“There's so much more to being a software engineer than typing the code, right? A lot of the work that you're going to do is, you know, you're going to be looking into a bug, You're going to be looking at the different files of the code base. Maybe you're going…”
Wu: Running parallel AI agents turns every engineer into an engineering manager
“One of the cool kind of features of Devon, I'd say, is, you know, if, as an engineer, you're working on, let's say, four different tasks today, you know, you just give one to Devon number one, you give the second one to Devon number two, the third one to Devon…”
Scott Wu: Software engineers spend 80-90% of time on implementation
“I think right now the average software engineer is probably spending like 10 or 20% of the time on that first thinking part
and they're spending 80 or 90% of the time on that implementation part.”
Scott Wu: There will be way more software engineers than ever
“And so, you know, I think software engineering, the job that we call software engineering is going to change, but I think practically, like, there's actually going to be way more software engineers than ever, you know, and I think there's a lot of precedent fo…”
Scott Wu: AI will make engineers 5-10x more effective
“I think, you know, our demand for software to be built is actually probably a lot more than 10 X what we're currently getting, and so, you know, I think what happens is we get to open up the power of software engineering to a lot more people, and every single …”
Devin runs full dev environments and servers to autonomously reproduce bugs
“In our code base, for example, Devon has all the setup that it needs. It has a machine that's basically instantiated where it can run the dev environment, it can run the server, it can run the front end, and so if it's, if you're asking it, hey, I need you to …”
Wu: AI automation will expose junior engineers to core architecture earlier
“You know, I think what happens, honestly, is I think that demand is going to just keep rising with supply. And I think the training process is going to change a little bit, but, you know, I think a lot of these core fundamentals of, you know, if you think of s…”
Ramachandran: AI flows will succeed AI agents in developer tooling
“Copilots are collaborative, but can only work on small scopes. Agents can do larger tasks, but are not collaborative. Both are very useful, but we realize that this real magic will happen when the AI has the ability to be both collaborative and independently p…”
Ramachandran: Cascade goes further than any other agentic system
“This allows Cascade to be independent, but Cascade takes it further than any other agentic system. By also generating commands to be run.”
Mohan: The Windsurf Editor is a direct fork of VS Code
“It is a fork of VS Code, and yes, We do know the memes. But there wasn't a need to rebuild everything from scratch.”
Mohan: Over 800,000 developers use Codeium's extension products
“We have over 800,000 sort of developers that use that product.”
Mohan: JetBrains has by far the best debugger for Java
“Very honestly, JetBrains has the best debugger for Java. It's not even close, right?”
Mohan: Coding is iterative exploration, not converting PRDs into output
“Code is not really built as you have a PRD and then you get some output out. It's more like you have a general vision and yes. And as you write the code, you get more and more clarity on approaches that don't work and do work. You're killing ideas and creating…”
Ramachandran: Capturing Editor Trajectories Lets AI Infer Intent Without Explicit Prompting
“If we actually understand that trajectory, then our ability for the AI to just be immediately be like, Oh, I understand your intent. This is what you want to do without you having to spell it all out for it. That is one like that kind of like magic would reall…”
Mohan: Codeium bypassed VS Code API limits by rendering Supercomplete via PNGs
“Problem was actually even to show the feature VS code would not expose an API for us to do this. So what we actually ended up doing was dynamically generating PNGs to actually go out and showcase this.”
Mohan: Stripping open-source commits turns coding agent evals into a continuous problem
“And if you do this across enough repositories, you've turned this highly discontinuous and discrete problem of make a PR work versus make it not work into a continuous problem. And now that's a hill you can actually climb.”
Mohan: Developers will never completely articulate problem statements to AI
“We believe that developers will never completely pose the problem statement, right? Because the problem statement lives in their head.”
Mohan: Windsurf's real value is large codebase work, not 0-to-1 apps
“We had the technology to go out and build these zero to one apps very quickly, and I think people are using Windsurf to actually do that, and it's like extremely impressive, but the real value, I think, is actually much deeper than that. It's actually that you…”
Ramachandran: SWE-bench and HumanEval do not reflect real professional software engineering
“Most evals and benchmarks that exist out there for software development is kind of bogus. There's not really a better way of putting it. Like, okay, you have SweeBench, that's cool, no actual Professional work looks like Sweebench, like human eval, same thing.”
Mohan: Squeezing the last 10% from AI benchmarks is counterproductive
“Once it starts hitting the peak of these benchmarks, getting that last 10% actually probably is, like, counterintuitive to the actual goal of what the benchmark was. Like, you probably should find a new hill to climb, rather than sort of p-hacking or really op…”
Ramachandran: Coding agents needing constant human fixes are unready for autonomy
“If the agent there has to always get human involvement to keep on fixing itself, it's probably not ready to become a full end-to-end automated system, because then we're just going to turn into a linter where, like, it produces a bunch of things and no one loo…”
Mohan: Explicit user prompting will soon become an anti-pattern in AI coding
“I actually think asking people to do things explicitly is probably going to be more of an anti-pattern if we can actually go and passively suggest the entire change for the user.”
Swix: Developers at large tech companies like Meta do not run code locally
“That's how it is at most companies, most like big calls, like Facebook, like nobody runs things locally.”
Mohan: Codeium uses in-house models for autocomplete due to poor frontier FIM
“The things like autocomplete and supercomplete that run on every keystroke are entirely, like, our own models, and by the way, that is still because properties like FIM, fill in the middle capabilities are still quite bad with the current model.”
Mohan: Vector embeddings fail at granular code retrieval queries
“For a lot of the systems, we do believe embeddings work, but for complex questions, We don't believe embeddings can encapsulate all the granularity of a particular query. Like imagine, imagine I have a question on a code base of find me all quadratic time algo…”
Mohan: Windsurf Cascade splits planning to Claude and codebase application internally
“The high level planning that is going on in the model is actually getting done with products like the cloud, but the extremely fast retrieval, as well as the ability to like take the high level plan and actually apply it to the code base is proprietary systems…”
Varun Mohan: Codeium does not pay third parties for indexing or custom model serving
“We don't really like pay other providers to do things for our indexing. We don't pay like other providers to do our serving of our own customer models, right?”
Varun Mohan: Codeium serves individual users and large enterprises with the exact same system
“When we're serving these products in an environment that works for these large enterprises, we're not going out and being like, we need to build this custom system for you guys, right? This is the same system that serves our entire user base.”
Mohan: Monetizing individual developers in AI tooling is the wrong strategy
“I think that right now optimizing for making money off of individual developers is probably the wrong, actually, strategy. Largely because I think individual developers can switch off of products, like, very quickly, and unless we have, like, a very large lead…”
Varun Mohan: Individual devs are price-sensitive; enterprises tolerate higher software costs
“Individual developers are much more sort of tuned towards small price changes. They care a lot more, right? Like if our product is 10, 20 bucks a month instead of 50 or a hundred bucks a month, that matters to them a lot. And for a large company where they're …”
Ramachandran: Codeium's $10/month plan primarily covers compute costs without high margins
“And so we just created a 10 dollar a month pro plan, mostly just to cover the cost. Like we're not really like operating, I think, on a much of a margin there either, but like, okay, like just to cover us there.”
Mohan: Microsoft Word spreads virally across companies through forced review compatibility
“Microsoft Word is a crazy product. It made it so that the moment you review anything in Microsoft Word, the only way you can review it is with other people in Microsoft Word. It's like this virus that penetrates everything, and it's not only penetrates it with…”
Varun Mohan: Parallel multi-agent coding trajectories don't require multiple machines
“And if you want to run all of them in parallel, you probably don't want end machines to go out and do it. I think that's unnecessary. Especially if most of them are I.O. Bound kind of operations where all you're doing is reading a little bit of data and writin…”
Ramachandran: Executing terminal commands in Cascade will be a huge unlock
“Vern's talked about not just suggesting the terminal command, but actually executing them. Like, I think that's gonna be a huge unlock.”
Mohan: Over 80% of software developers are on Windows
“A lot of people, once again, over 80% of developers are on Windows.”
Ramachandran: Codeium grew from zero to $10M ARR in under a year
“I don't think we go from zero to ten million ARR in less than a year in an enterprise product if we don't have a product that developers love.”
Mohan: Senior engineers gained little value from ChatGPT due to context friction
“I don't think most of us got much value from Chachapi team. Largely because I think the problem was, and this is maybe a little bit of a different thing, it's, like, a lot of the engineers at the company who have been writing software for, like, over eight yea…”
Mohan: Codeium will never launch a product with a waitlist
“We, by the way, we will never launch with a waitlist. We will never launch with a waitlist.”
Mohan: Supercomplete improved materially from user trajectory and synthetic data
“The super complete product that we have has gone materially better because of us not only using synthetic data, but also getting the preference data from our users of, like, hey, given these set of trajectories,
Here's actually what a good outcome is.”
Mohan: Post-acceptance code edits provide better training signal than mere acceptance rates
“One of the really beautiful parts about our product that is very different than a chat GPT is we can not only see if the acceptance happened, but if something more than the acceptance happened and it happened even more than that, right? Like, let's say you acc…”
Mohan: In-terminal AI commands were impossible to build in VS Code
“We literally couldn't do that. In VS Code.”
Ramachandran: Codeium is currently undergoing FedRAMP accreditation for defense clients
“Right now, you know, we're, have a lot of customers in, like, the defense space, for example. We're going through, you know, FedRAMP accreditation right now”
Mohan: Buying third-party software risks permanently losing core engineering competencies
“When you go out and buy, you're losing a core competency inside the company. And that's a core competency you can never get. It's, or it's very hard. Like, startups are so limited on time.”
Mohan: Codeium would not have survived if it relied on vLLM
“If we use VLLM, we would not be talking with you right now.”
Mohan: Scale AI increased revenue 10x with each major pivot
“Effectively, it starts out as, like, you know, human task interface, then an AV labeling company, then a cataloging company, then Now a generative AI labeling company, and every time the revenue of the company kind of goes up by a factor of 10, even though the…”
Mohan: Codeium closed 30 to 40 customers before hiring sales VP
“Yeah, we had sold the product to like, 30 or 40 customers of that kind.”
Hou: Codeium garnered a couple million users across editor extensions
“So we garnered a couple million users on our VS Code, JetBrains, Vim, Emacs extensions.”
Hou: Windsurf generated 4.5 billion lines of code in three months
“So in three months, we've been generating 4.5 billion lines of code.”
Hou: Windsurf is among Anthropic and OpenAI's largest consumers
“We've had immense success getting people onto the platform, and we've been very fortunate to have the issue of being some of Anthropic and OpenAI's largest consumers.”
Hou: The future of developer tooling will eliminate copy-pasting into AI
“So we very strongly believe in a future of no copy-paste. Right? You should never have a situation where you're in a terminal, or you're in a document, or even on a website, and you're copy-pasting text into an agent. That's just not how the way the world work…”
Hou: Future AI agents will anticipate developer actions dozens of steps ahead
“In the future, we'll be looking not just one to five steps in the future, but 10, 20, 30 steps into the future. It'll be writing unit tests before you've even finished defining the function. It'll be performing code-based-wide refactors on multiple files based…”
Hou: 99% of AI editor rules file contents will be automatically inferred
“We strongly believe that having a rules file, you know, we do allow users to add a rules file, we strongly believe that a rules file is a crutch. You know, by the end of twenty-twenty-five, 99% of the things that you're gonna put in a rules file will be interp…”
Hou: Chat interfaces are a legacy paradigm being replaced by AI agents
“We only have an agent, and it's called Cascade inside of Windsurf. Chat is a legacy paradigm, and we completely replaced it, and as you can see here, users are enjoying it, or in fact, they might not even know the difference, but they're just enjoying the high…”
Hou: Developers will no longer need explicit @-mentions in AI IDEs
“In the long term, we believe that LLMs are going to improve, and they already have improved to the point where you don't need to explicitly specify an at mention. The LLMs should be intelligent enough to pick it up.”
Hou: 90% of code written by Windsurf users is generated by Cascade
“All of our users, 90% of the code that they're writing is generated with Cascade. That's an astonishing number. Autocomplete was more in like the 20, 30%.”
Wu: Autonomous coding agent capability currently doubles every 70 days
“What you see in general is that that doubling time is about every seven months, which already is pretty crazy, actually, but in code, it's actually even faster. It's every 70 days, which is two or three months, and so, you know, if you look at various software…”
Wu: Tab completion was the only AI coding product with PMF 18 months ago
“18 months ago, I would say, the only really, the only product experience that had PMF in code was just tab completion, right? It was just like, here's what I have so far, predict the next line for me. That was kind of all you really could do in a way that real…”
Wu predicts AI coding agents will advance 16x to 64x in 12 months
“And I think that, you know, we're gonna see another 16 to 64 X over the next 12 months as well.”
Windsurf aims to shift coding workflows from 80% agent to 99% agent
“And today, timelines are 80 to 90% agent, 10 to 20% human. But we're trying to build towards a future that gets the 99% agent and one percent human. We only want to ask the user for final approval.”
Hou: Windsurf's SWE-1 achieves near-frontier model results at lower cost
“And we've been able to achieve near frontier model results at the fraction of the cost, and with a significantly smaller team.”
Hou: Users choose SWE-1 at higher frequencies than Claude 3.7 and 3.5
“People are choosing SWE-ONE because it recognizes how they do work, not necessarily how to generate code. And it's contributing, actually, an even higher frequency than models like 3.7 and 3.5.”
Wang: Other companies and VCs sought deals with Windsurf before Cognition merger
“When Scott, when I talked to Scott, it was just like, Hey, this kind of makes sense. Like there was other companies. That were interested. There was actually VCs that wanted to invest. But Scott, as a person with Russell, and then I was with Graham just talkin…”
Wang: Windsurf employees received 10 to 12 poaching emails during merger weekend
“And I don't think it was reported, but like time was not our friend. We had customers like starting to email like, Hey, what's the, what's going to happen to the company. And we had recruiters. I think almost every employee probably got like 10 or 12 recruitin…”
Wu: Cognition has no plans to change Devin and Windsurf product philosophies
“What I would say at a high level is, look, I think spiritually, you know, we really want to maintain the philosophies of both of these products and we have no plans to change that.”
Wang: Prior to merger, Devin was building an IDE and Windsurf an agent
“We actually looked at each other's roadmaps, like product roadmaps and the Devon team was actually looking to build an ID and our team was actually looking to build a remote agent.”
Wu: Global software engineers grew from under 1M in 2000 to over 30M in 2025
“In the year 2000, there were less than a million software engineers in the world. And, you know, this was whole dot com era, like, you know, a lot of software was being built. But now in the year 2025 today, there's actually over thirty million software engine…”
Scott Wu: No customer or training data was shared in Windsurf transaction
“There was no, no, no information that was given out, for example, in terms of customer data, training data, any things like that. And so, so, you know, all of that is, is, is strictly proprietary and then remains, you know, our exclusive access.”