DeepCode v2 turns coding-agent work into a loop

Chao Huang says DeepCode v2 keeps working on a natural-language goal through repository understanding, implementation, tests, verification, and repair. The release also adds project-specific automations that can run manually or on a schedule, plus shared project and session history across its CLI and desktop app. Why it matters: The useful distinction is not whether an agent can write code once. It is whether the system can continue after the first change, run checks, preserve the work already done, and give the operator a way to pause or redirect it.

Try this: For one coding-agent task, write the finish line before you start: the files to change, checks to run, evidence required, and the point where a person must review the result. Then see whether the agent can complete that loop without repeated prompting.