Coding agents sped up code generation while specification and verification remain the bottleneck

Farley, writing up his AI Native DevCon London talk, argues that coding agents have sped up generating code without improving the practices that make the result durable: specification, verification, and incremental delivery. "So the bottleneck is not code generation. We have sped up the coding bit, which was often not the hardest part. The bottleneck is specification, verification, and release," he writes. His proposed fix is to write acceptance criteria as executable, version-controlled specifications, kept mostly out of the model's view, and to treat tests as a form of measurement rather than proof that the work is correct.

Why it matters: Teams already running agents day to day can write executable specifications that test the agent's output and keep continuous delivery's incremental habits intact around it.