What ClickHouse learned from a year of coding with AI agents

ClickHouse says AI coding agents went from disappointing to genuinely useful over the past year.

Earlier agents struggled on its large C++ codebase, but newer models like Claude Opus 4.5 made daily use practical.

The biggest wins are repetitive code changes, merge conflicts, code review, bug investigation, and flaky-test cleanup.

ClickHouse also built custom tooling and some autonomous agents to catch edge cases and support CI.