Inside Devin: The AI engineer that's set to write 50% of its company’s code this year | Scott Wu
Watch on YouTube ↗ Summary based on the YouTube transcript and episode description. Prompt input used 79979 of 112555 transcript characters.
Cognition CEO Scott Wu says Devin already writes 25% of company PRs and will exceed 50% by year-end on a 15-person team.
- Devin merges several hundred PRs/month at Cognition; currently ~25% of all PRs, targeting >50% by end of 2025.
- Each of Cognition’s 15 engineers runs up to 5 Devins concurrently, treating them as async junior teammates.
- Cognition went through ~8 pivots within the coding-agent space before landing on the current Devin product form.
- Co-founder Walden was an early Cursor engineer; co-founder Steven was an early Scale AI engineer — team met via competitive programming.
- Wu argues AI lacks hardware-distribution friction unlike PC/internet/mobile, making its growth curve far more explosive.
- Devin can spawn child Devin instances via API, enabling parallelized subtask execution within a shared context.
- Wu frames the engineer role shift as bricklayer-to-architect: the interesting 10% (architecture, problem definition) expands; the 90% (debugging, migrations, bug fixes) gets delegated.
- To hire a key MIT junior, Wu flew to North Carolina and had dinner with the candidate’s parents to negotiate a school-plus-work arrangement.
2025-05-04 · Watch on YouTube