The PodKnowledge|← All Episodes
Based on Lenny's Podcast data
Lenny's Knowledge Sketch

Your Junior Buddy,
Named Devin

Scott Wu
Co-founder & CEO, Cognition · Makers of Devin, the first autonomous AI software engineer · 8 pivots in 18 months to land here
SEP 8 2025
The Product

A Junior Buddy
You Tag in Slack

@ IN SLACK/LINEARDEVIN AGENTPR ON GITHUB
"Devin is your junior buddy. You tag it in Slack, in Linear, on an issue, and Devin makes pull requests in your GitHub."
  • A fully asynchronous workflow — not a text-completion tool, not a chat window with a copy button
  • Devin's seniority is a moving target: high-school CS student at launch, then college intern, now a junior engineer
  • "Jagged intelligence" — better than a human at some things, worse at others, and the shape keeps shifting
  • Customers range from one- or two-person startups to Fortune 100s and public banks
Framework

Cognition,
Building Devin With Devin

15 ENGINEERS5 DEVINS EACH100s of PRs / mo
15
engineers on the Cognition team
25%
of Cognition's PRs today are Devin's
50%+
the target by end of the year
  • Every engineer on the team is working with up to 5 Devins in parallel
  • Devin merges "several hundred pull requests" into production in Cognition's codebases every month
  • Sometimes Devin does the task 100% autonomously; sometimes the human jumps in for the last 10-20%
  • The learning loop is deliberate: Devin accumulates knowledge from every engineer, brings it into new sessions
Scott's read on the change"Programming is only going to become more and more important as AI gets more powerful." The role shifts from writing code to instructing the computer.
The Backstory

Eight Pivots to Land Here

  • Then: Hacker houses full of iterations. The core belief was always coding + agents
  • Pivots: eight of them in a year and a half, all within coding agents
  • The unlock: moving from text-to-text to an actual autonomous system that iterates, tests, decides
  • The naming pattern: Devin (Cognition), Cursor (Anysphere), Bolt (StackBlitz), v0 (Vercel). The winning product almost always shows up later in the company's life, with its own name.
Why treat Devin as a person

Naming the agent, giving it a personality, was a deliberate design choice. Scott says it "captures the soul" of an entity you hand off tasks to, teach, and learn with over time. Nothing else on the market does this.

Product changes since launch

At launch, no Slack, no GitHub integration, no Linear, no interactive planning phase. Users would just get finished code with no way to steer. Almost every product primitive today came from watching customers.

Playbook

How Scott Onboards a Team

  • Do NOT start with "re-architect the whole codebase." Users try this constantly. It fails constantly.
  • Get the repository set up first: virtual machine, tests, linter, CI — things Devin needs to actually run code
  • Hand Devin a couple of one-pointer tasks so it can get familiar with the codebase
  • Then move to async multi-Devin mode: kick off runs, only jump in to steer or give feedback
The 80/90% patternDevin's not perfect. The winning workflow is scope-then-let-go, then come in at the end to review and give feedback on the last 10-20%. Sometimes you check mid-flight to course-correct. Usually you don't.
Contrarian

Scott's Contrarian Reads

✗Rate agents on one benchmarkINSTEAD →✓ Jagged intelligence — better than human at some things, worse at others, and the shape keeps shifting. Any single "level" comparison misleads.
✗AI will kill engineering hiringINSTEAD →✓ Scott's belief: hiring goes up. Instructing computers is more valuable, not less, when the computer is more capable.
✗Ship the model, users will figure out the UXINSTEAD →✓ A huge fraction of Cognition's last year went into UX — Slack tags, Linear integration, planning phases, mid-flight touch-ups — not just capability.
✗Agents mean less human context to teachINSTEAD →✓ The team teaches Devin like they'd teach a new hire — because Devin brings the accumulated knowledge into every new session, permanently.
Based on Scott Wu's episode on Lenny's Podcast. All ideas on this page are from the episode.Watch on YouTubeFollow @scottwu46 on X
GO DEEPER IN THE EPISODE
09:13 Scaling and future prospects10:23 Devin's origin story17:26 The idea of Devin as a person22:19 How a team of “Devins” are already producing 25% of Cognition’s pull requests25:17 Important skills in the AI era30:21 How Cognition’s engineering team works with Devin's34:37 Live demo42:20 Devin’s codebase integration44:50 Automation with Linear46:53 What Devin does best52:56 The future of AI in software engineering57:13 Moats and stickiness in AI01:01:57 The tech that enables Devin01:04:14 AI will be the biggest technology shift of our lives01:07:25 Adopting Devin in your company01:15:13 Startup wisdom and hiring practices01:22:32 Lightning round and final thoughts
Every sketch here began as a transcript.

Send us yours and we’ll draw it. $1 a transcript.

Sketch mine →
𝕏︎ X / Twitterin LinkedIn📸 Instagram🔗 Copy link
0:00