← back
Devin 2.0 and the Future of SWE - Scott Wu, Cognition
Takeaway
Coding agents are scaling on a 70-day doubling curve; product strategy must re-baseline every model generation as Devin moves from migrations to multi-file feature work.
Summary
- AI-agent capability in code doubles every ~70 days vs ~7 months for general agents, implying 16–64× growth in task length per year.
- Tier 1 (mid-2024): repetitive migrations (JS→TS, Angular upgrades) solved via Devin's 'playbooks' for reliable step-following plus persistent memory.
- Tier 2 (late 2024): isolated intern-level bugs/features requiring repo setup, lint/CI, snapshot/rollback VMs.
- Tier 3: cross-file bugs needing code-as-hierarchy reasoning rather than text-only context.
- Devin 2.0 stresses that the right interface, bottleneck, and capability shift every 2–3 months alongside models.
devincoding-agentscognition
Original description
A talk on the future of software engineering with Scott Wu of Cognition AI, the makers of Devin. About Scott Wu Scott is the co-founder and CEO of Cognition AI. He previously competed in international programming competitions (3x IOI gold medalist) and co-founded Lunchclub, an AI-powered professional networking platform. Scott grew up in Louisiana and attended Harvard University. Recorded at the AI Engineer World's Fair in San Francisco. Stay up to date on our upcoming events and content by joining our newsletter here: https://www.ai.engineer/newsletter