- Joined
- Mar 2026
- How they joined
- Directly
- Before Cognition
- VercelStaff Software Engineer
- Status
- Current
“on the office vibes tier ranking cog would be S+++”
- Joined Cognition around late Feb 2026; in his May 29, 2026 Cognition blog post he wrote verbatim: "3 months ago, I joined Cognition to help build the future of software engineering. Devin has come a long way since launching as the first AI software engineer, and I've been blown away watching the team behind it actually use Devin every day."May 29, 2026Verifying Agentic Development at Scale, by Ido Pesok: '3 months ago, I joined Cognition to help build the future of software engineering.'
- At Cognition he works on autonomous end-to-end verification for Devin: Devin uses its own cloud computer to run the app, exercise flows and return a chaptered test video with pass/fail assertions. He wrote: "From validating our Slack integration to testing complex Windsurf features, the team always has an army of Devins in test mode.".May 29, 2026Verifying Agentic Development at Scale, by Ido Pesok: '3 months ago, I joined Cognition to help build the future of software engineering.'
- When Cognition announced it was acquiring Corridor (Oct 29, 2026), he reposted it and wrote, verbatim: "Super excited to get to work with @jackhcable + @AshwinRamaswami and the amazing Corridor team.".Oct 29, 2026Super excited to get to work with @jackhcable + @AshwinRamaswami and the amazing Corridor team.
- Sept 16, 2026 X post, verbatim: "We have seen a lot of success with Agentic MapReduce in helping engineering teams find vulnerabilities across their codebase. We are bringing that same agent swarm ability to any Devin session with “/scan”.".Sep 16, 2026We have seen a lot of success with Agentic MapReduce in helping engineering teams find vulnerabilities across their codebase. We are bringing that same agent swarm ability to any Devin session with “/scan”. Do automated codebase cleanup, DB performance checks, and so much more! Excited to see teams try this and hit us up if you have any feedback!
- July 30, 2026 X post on Devin Review stacked PRs, verbatim: "Coding has always been extremely fun for me, but as I was transitioning to no longer coding by hand, I found myself missing that flow state. The PRs agents made were too large to merge and often littered with a ton of small bugs. Stacked PRs have brought the flow state back." He credits @dkrajews123, @sahinolut and @jaredpalmer.Jul 30, 2026Coding has always been extremely fun for me, but as I was transitioning to no longer coding by hand, I found myself missing that flow state. The PRs agents made were too large to merge and often littered with a ton of small bugs. Stacked PRs have brought the flow state back. Rather than getting stuck in endless prompting loops, I feel like I’m constantly moving the needle on what I’m working on. With first party support of stacked PRs in Devin Review, Devin can autonomously iterate on concentrated bugs in each section of your stack. When you come back to the session, you can review digestib
- Before Cognition he was at Vercel from 2024 to 2026 on the v0 team, rising from Software Engineer to Senior to Staff Software Engineer (2026). He left as Staff.2026
- The blog post says test runs approved per day on Devin "have more than doubled" in the past couple of months. It names known failure modes he is working on: screenshot timing (missed toasts) and models 'cheating' by running JavaScript instead of clicking through the UI.May 29, 2026Verifying Agentic Development at Scale, by Ido Pesok: '3 months ago, I joined Cognition to help build the future of software engineering.'
- A Cognition sales post credits "Ido Pesok, Nicholas W., Angela L. (here) & team" for the Devin security work (Devin Security Swarm, 94% security-risk reduction at a financial-services firm).2026
- Gave the talk 'Evals Are Not Unit Tests' (as Vercel v0 engineer) at AI Engineer World's Fair 2025. It argues that nondeterministic LLM apps need systematic evals built from production logs and feedback.Jun 2025
- At Vercel he co-authored 'Eval-driven development: Build better AI faster' with CTO Malte Ubl and Alice Alexandra Moore. He also helped explain v0's composite model architecture.2025
- Open source: created ZSA (IdoPesok/zsa), a type-safe Next.js server actions library with schema validation, middleware, retries and React Query integration. Profile data shows 143 PRs and 274 pushes from May 2024 to Jan 2025.May 2024
- Open source: led energizeai/actions, described as 'the world's best open source registry of LLM tools', with 110 PRs from Feb to Apr 2024.Feb 2024
Led development of the v0 agent harness
Pushed the frontier of Next.js code gen quality and evals
Helped launch https://v0.app
Led frontend and LLM development.
Worked on payments export migration.
Exploring, discussing, and contributing to breakthroughs in AI with fellow students from Cal Poly SLO.
Directed frontend for pension tracker app built to support firefighters and police. Onboarded by LAPD, SFPD, and more.
Graduated in 3 Years
Learned a lot about AI. Built 12 projects revolving around optimization, neural networks, natural language processing, and more. Video attached displays the convolutional neural network project. Requirement was to train a Sequential model to predict traffic signs given an input image, using the Keras API.
Startup School is an online program for founders actively pursuing their own startup, hosted by Y Combinator.
Verifying Agentic Development at Scale, by Ido Pesok: '3 months ago, I joined Cognition to help build the future of software engineering.'
Super excited to get to work with @jackhcable + @AshwinRamaswami and the amazing Corridor team.
Coding has always been extremely fun for me, but as I was transitioning to no longer coding by hand, I found myself missing that flow state. The PRs agents made were too large to merge and often littered with a ton of small bugs. Stacked PRs have brought the flow state back. Rather than getting stuck in endless prompting loops, I feel like I’m constantly moving the needle on what I’m working on. With first party support of stacked PRs in Devin Review, Devin can autonomously iterate on concentrated bugs in each section of your stack. When you come back to the session, you can review digestib
We have seen a lot of success with Agentic MapReduce in helping engineering teams find vulnerabilities across their codebase. We are bringing that same agent swarm ability to any Devin session with “/scan”. Do automated codebase cleanup, DB performance checks, and so much more! Excited to see teams try this and hit us up if you have any feedback!
excited to share how we work with Devin on Slack!
on the office vibes tier ranking cog would be S+++
. @penlume 🐐
