- Joined
- Nov 2023
- How they joined
- Directly
- Before Cognition
- CursorEarly Engineer
- Olympiads
- IOI gold 2020IOI 2020Gold
“I got sick of reviewing code. So I made /walden-review. Its a skill auto-generated against my last 6 months of Github comments and agent sessions. It learned what I look for and even my comment style: “blunt, occasionally profane”.”
- In a May 28, 2026 interview, Walden Yan explained Cognition's decision to separate Devin's "brain" from its development machine.May 28, 2026
- Walden Yan explained the team's choice of full virtual machines through the need to run and test real applications, including applications that themselves use Docker.May 28, 2026
- Walden Yan treated testing as a broader reasoning problem than clicking interface coordinates.May 28, 2026
- Walden Yan reported cases in which Cognition had to orchestrate multiple frontier models to complete testing work.May 28, 2026
- Walden Yan said he had added annotations to testing videos so viewers could understand what Devin was demonstrating.May 28, 2026
- Walden Yan described the team tuning Devin's GitHub interactions so it could receive review comments, address useful findings, and push back when a comment was wrong.May 28, 2026
- Walden Yan credited Cognition's infrastructure colleagues with diagnosing slow grep searches as network-backed filesystem overhead, and the team replaced that filesystem.May 28, 2026
- Walden Yan distinguished the grep filesystem fix from the separate BlockDev storage format, which reduced VM save-and-restore work by processing filesystem changes rather than the entire disk.May 28, 2026
- By the May interview, Walden Yan said the described storage system had already been superseded internally.May 28, 2026
- Walden Yan recounted internal experiments building intended products through successive AI changes with automatic merging and no code review.May 28, 2026
- Walden Yan estimated that the December state of the art sustained this approach for about two weeks before maintenance became troublesome.May 28, 2026
- Walden Yan's example was a button implemented in ten places with inconsistent variations, making a color change difficult.May 28, 2026
Building Cursor (www.cursor.so)
Next-generation security through automated program verification.
Sourced leads, built web applications and websites, managed digital media presence for clients, managed a team of 14 consultants and engineers. Got acquired
Conducted research in cryptography and machine learning. Wrote papers and presented them to peers. Designed and benchmarked NPS models, beating state-of-the-art. Collaborated on a compiler library for optimized fully homomorphically encrypted programs integrated with LLVM.
I was just reflecting on the moment we realized Devin had the potential to become the world’s first AI software engineer. This was a few years ago, when AI was just a good autocomplete in your IDE. But I’d find myself waking up every day wishing I could text an agent and have it magically build my projects for me. One day, the early team (then a group of hackers, not a company) was really stuck trying to spin up a MongoDB server. The docs were really hard to read. I'd been at it for a while when I finally turned to our very rough Devin prototype and said, “hey, can you get the server work
Devin usage since late last year is up more than 7x. A lot of that is driven by Devin AI just choosing to run itself, a lot more. With new feature development, AI agents often still need lots of input from human engineers. But a lot of software engineering isn't shipping new features - it's maintenance and fixing bugs. And for those use cases, Devin now often automatically responds to reports, does deep investigation by itself, ships fixes, tests them and gathers evidence that the fixes work. It sometimes even spins up a whole team of other Devins, deploying its own engineering team t
We shipped Windows VM support for Devin today. We're the first cloud agent with this capability. This was a surprisingly deep infrastructure problem. Every layer of our Devin Cloud stack assumed Linux VMs. Hypervisors, secrets management, shell infrastructure. We had to rethink each one from the ground up. The result: Devin can build, run, and verify .NET codebases and Windows applications natively. Check it out here: http://devin.ai/windows
Devin now works like a full time junior support engineer on your team. You don't even have to @ tag it with prompts. With Auto-Triage, you can have a Devin proactively monitor an entire Slack channel for incidents, bug reports, or feedback. And that Triage Devin decides if it needs to @ tag a sub-Devin to investigate. The sub-Devins running in parallel look through your analytics, search your codebase, reference related tickets and threads, and come back with fix for you to approve. And because the main Triage Devin maintains a long-running context of all the issues its ever looked at a
Everyone’s been talking about how Claude Mythos found thousands of security flaws. But here’s the bigger problem: who’s going to fix them? Introducing Devin for Security - a new set of workflows purpose-built for security teams to use autonomous coding agents to remediate issues. Because the hard problem in cybersecurity isn’t just finding weaknesses in your system. It’s fixing them. In fact, security alerts are already abundant. An issue might be flagged on Monday. The security team opens a ticket. Engineering plans the change. By Friday, the code is still waiting for review, and an attack
New blog about building multi-agents: A year ago, I'd tell people to not build multi-agents and to focus on context engineering fundamentals. Today, many sexy ideas are still impractical, but we've found some setups that actually work https://lnkd.in/gHpyza7P
linkedin.com/posts/waldenyan_cognition-multi-agents-whats-actually-activity-7453125293327929344-AfCe@DevinAI solve teleportation. Feel free to use subagents
@BrendanFoody @adarsh_exe @suryamidha are some of the hardest working founders I know. Incredibly motivating to watch them succeed. They're doing some very cool stuff with Devin 🫡
The bar for quality in FC 2.0 is very high We rejected 80% of the PRs submitted and sent them back for OSS maintainers to improve This is now the eval that best matches my vibe check of models
Devin Desktop is actually good now https://t.co/Cu2F3QSoxD
I got sick of reviewing code. So I made /walden-review. Its a skill auto-generated against my last 6 months of Github comments and agent sessions. It learned what I look for and even my comment style: “blunt, occasionally profane”. I always run it before ever reading a PR myself, and it catches design/codepattern/architectural problems most bug-catcher miss. FYI: This method burns a lot of tokens First, I asked Devin to collect my last 6 months of GitHub PRs and Devin sessions, collect every comment I left or message I sent, filter out unimportant stuff (“lgtm”), and build a hillclimbing se
The easiest way to get more than $29/mo of usage will be to work somewhere else Eng leaders will need to learn to train their engineers how to get ROI on token usage, not limit it
