Skip to content
Lenny's PodcastLenny's Podcast

Scott Wu: Why Devin will write half of Cognition's code

Cognition's 15 engineers each run up to five Devins in parallel; a quarter of monthly PRs already ship from agents, freeing architects to scope tickets.

Scott WuguestLenny Rachitskyhost
May 3, 20251h 32mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

Inside Devin: How AI Engineers Are Reshaping Software Teams and Workflows

  1. Cognition Labs CEO Scott Wu explains Devin, an autonomous AI software engineer that companies use like a junior remote developer via Slack, Linear, and GitHub.
  2. Within Cognition’s 15-person engineering team, each engineer typically runs around five Devins in parallel, and roughly a quarter of their monthly pull requests are authored by Devin—expected to exceed 50% by year-end.
  3. Wu argues AI will shift engineers from “bricklayers” to “architects,” expanding, not shrinking, the total number of programmers by making software creation far more efficient and accessible.
  4. The conversation also covers Devin’s origin story and pivots, its agentic product design, how it learns a codebase and supports onboarding, and broader implications of AI’s explosive, hardware‑unconstrained adoption curve.

IDEAS WORTH REMEMBERING

5 ideas

Treat Devin like a junior engineer, not a smarter autocomplete.

Wu stresses that users get the most value when they give Devin well-scoped tasks (tickets, bugs, small features) and collaborate through reviews and clarifications, just as they would with a human junior teammate.

Run multiple agents asynchronously to massively increase throughput.

At Cognition, each engineer commonly runs up to five Devins in parallel, handing off different issues and features so they can focus on higher‑level design and review instead of serial implementation work.

Start with simple, verifiable tasks to onboard Devin into your codebase.

Successful teams first let Devin handle one‑pointers—small fixes, UI tweaks, tests, documentation—while teaching it how to run CI, linting, and local tests, then gradually scale to more complex projects.

Engineers will shift from ‘bricklayers’ to ‘architects,’ but coding literacy remains crucial.

Wu argues people should absolutely still learn to code: understanding abstractions, systems, and trade‑offs is what lets humans specify what to build, peel back layers when needed, and fully leverage AI capabilities.

AI will likely increase, not decrease, demand for software engineers.

Invoking Jevons paradox, Wu predicts that as the cost and friction of building software drop, society will find far more software to build, expanding the total amount of code and the number of people involved.

WORDS WORTH SAVING

5 quotes

Devin is a fully autonomous software engineer that is gonna work on tasks end-to-end.

— Scott Wu

Our whole team is only like 15 engineers. We use a ton of Devin when we're building Devin.

— Scott Wu

I really think that programming is only going to become more and more important as AI gets more powerful.

— Scott Wu

One of the ways that we've kind of thought about Devin is really allowing engineers to go from bricklayer to architect.

— Scott Wu

As it becomes easier and easier to program, I think we're gonna have a lot more programmers, not fewer.

— Scott Wu

What Devin is and how it works as an autonomous AI engineerHow Cognition’s own team uses Devin (5+ agents per engineer, PR stats)The evolving role of software engineers: from implementers to architectsProduct and UX design for agents versus traditional chatbots or IDE toolsTechnical and strategic bets: reinforcement learning, agents, and code as a domainAI industry landscape, competition, and thoughts on moats/stickinessPractical adoption patterns and best practices for teams deploying Devin

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome