How Devin replaces your junior engineers with infinite AI interns that never sleep

Scott Wu, co-founder and CEO of Cognition, discusses Devin, an autonomous AI software engineer that works asynchronously like a junior remote engineer. The conversation covers how Cognition's 15-person team uses multiple Devins to build Devin, how engineering roles may shift from bricklayer to architect, the company's pivots and origin story, and lessons on stickiness, hiring, and startup execution.

01Key learnings

• Treat an AI agent like a junior engineer: scope it with well-defined tasks rather than open-ended problems, and start with small, easily verifiable tickets.
• Run several agents asynchronously in parallel and only step in for the portions needing human judgment, such as scoping, architecture, or final review.
• Invest up front in setup for the agent: connect repositories, teach it how to run lint and CI, and give it a virtual machine so it can test its own code.
• Expect engineering work to shift from writing boilerplate and debugging toward defining problems and architecting solutions, while still learning fundamentals to understand abstractions.
• Think about stickiness rather than moats: agents accumulate codebase knowledge and team context over time, making switching costs grow with usage.
• Fundamentals still matter: understanding how databases, networking, and abstractions work helps engineers direct AI tools precisely and peel back layers when needed.
• Focus on the core startup fundamentals, such as building something people want, staying close to customers, moving fast, and hiring great people, and push them harder than seems reasonable.
• Design for where the technology is heading in the next few years rather than only current capabilities, since AI capabilities change quickly.

02Frameworks mentioned

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.