Devin is billed as an autonomous AI software engineer

Cognition unveils Devin, an agent that plans and executes multi-step software tasks, resolving ~14% of SWE-bench issues.

What happened and why it matters

Devin popularised the idea of an end-to-end coding agent with its own shell, editor and browser. Its demos were later criticised as overstated, but the 13.9% unassisted SWE-bench result was a jump over prior systems.

Sources

• Introducing Devin, the first AI software engineer ↗ Cognition · 12 Mar 2024