The headline on Claire Vo’s How I AI session with Scott Wu is engineered to provoke: “How Devin replaces your junior engineers with infinite AI interns that never sleep.” The episode itself — a screen-share of how Cognition uses its own product — argues something more specific and more defensible. It resurfaced this summer as Wu made the rounds: a Master Stage keynote at RAISE Summit 2026 in July (“The Golden Age of Software Engineering,” one of his first major European appearances) and a Sourcery interview that describes Cognition as the $26 billion applied AI lab behind Devin and Devin Desktop.
What the demo actually shows
Strip the headline and the episode is a workflow documentary. The published rundown lists what Wu walks through:
- DeepWiki as the research step — query the codebase, then write a better-scoped prompt from what you learned. (The DeepWiki docs describe the same loop: architecture answers with linked sources, feeding session prompts.)
- Asynchronous delegation — treat Devin like a team member you hand scoped tasks to while you’re in meetings, not a pair-programmer you babysit.
- Public Slack channels over DMs — both humans and the agent learn from watching solutions land in the open. (Slack integration docs cover the mechanics.)
- A “first line of defense” pattern — the agent analyzes crashes and incidents before a human is paged.
- His top-five task list — frontend fixes, version upgrades, documentation, incident response, and testing.
That last list is the tell. Those are not junior-engineer jobs — they are the toil fraction of every engineer’s job. The “infinite interns” framing is about adding elastic capacity for that fraction, which matches Cognition’s own guidance in When to Use Devin: well-scoped, verifiable tasks first; judgment stays with people.
The capacity thesis has a paper trail
The framing is consistent with where Cognition has put money. The TierZero acquisition extends Devin toward production operations — incident response as a product surface, not a demo pattern. And the demand-side arithmetic Wu keeps gesturing at (agents make PR creation cheap, which moves the constraint to review and merge) is the same one our earlier desk piece traced through Devin Review as the new rate limit: capacity added upstream reappears as queue pressure downstream.
That is the honest version of “infinite interns.” Interns generate work-in-progress; someone still owns what ships. The episode’s most transferable advice — small scoped tasks, context in public, agents on the pager before humans — is really a description of how to be a good manager of that queue.
Method: Episode content is drawn from the published episode page and chapter list (fetched 2026-08-12), cross-checked against Cognition's docs for product behavior. Valuation figure ($26B) is as stated in the Sourcery episode description. Interpretation is labeled as ours.
References
- How Devin replaces your junior engineers with infinite AI interns that never sleep — How I AI with Claire Vo; workflow rundown and task list.
- The Golden Age of Software Engineering — Scott Wu, RAISE Summit 2026 — July keynote.
- Inside the Fastest-Growing Category in AI — Sourcery interview; $26B description.
- When to Use Devin — official scoping guidance.
- DeepWiki docs — research-then-prompt loop.
- Slack integration docs — channel-based delegation.
- Welcoming TierZero to Cognition — operations extension.
Editorial from Devin Central — a fan news desk, not Cognition. Devin is a trademark of Cognition.