Chunk sidecars launched for AI agents

Compared to the fast code generation speed of AI agents, the existing CI validation method has the problem that verification happens after code is pushed, resulting in a slow feedback loop and making it easy for agents to lose task context.

Chunk sidecars solve this problem by running scoped microbuilds in an environment that mirrors the actual CI environment.

The key features are as follows:

• Fast feedback: Validation is completed in around 27 seconds on average, much faster than the existing method where a full build takes about 5 minutes.
• Cost and efficiency: Token usage from retry loops can be reduced by 3 to 5 times.
• High compatibility: It can be used with Claude Code, Codex, Cursor, and custom agents.

This feature is provided free of charge to all CircleCI users.