ClickHouse Embraces Agentic Coding: Practical Applications Beyond the Hype
Key Takeaways
- ▸ClickHouse signed contracts with multiple agentic coding providers (Anthropic, Windsurf, Cursor) after resolving legal and security concerns
- ▸Coding agents prove most valuable for boilerplate tasks, performance tests, dashboards, and internal tools rather than core C++ development
- ▸Recent model improvements (Claude Sonnet 4.5, Claude Code) have substantially enhanced coding agent capabilities compared to earlier versions
Summary
ClickHouse has adopted agentic coding tools from Anthropic, Windsurf, and Cursor to enhance developer productivity, moving beyond theoretical discussions to practical implementation. The company found that while coding agents excel at boilerplate tasks, small scripts, and internal tools, they remain challenging for complex C++ backend development. ClickHouse developed its own custom agents (DWAINE, CAISER, and TRAISA) and invested in observability tools like Librechat and Langfuse to manage AI-assisted development workflows. According to the company's analysis, recent improvements in models like Claude Sonnet 4.5 have significantly improved agent capabilities, though skepticism remains about their utility for complex, mission-critical codebases.
- ClickHouse developed custom agents and observability infrastructure to manage AI-assisted development at scale
Editorial Opinion
ClickHouse's pragmatic approach to agentic coding offers a refreshing counterpoint to both AI evangelism and blanket skepticism. By treating coding agents as specialized tools for specific tasks rather than universal replacements, the company demonstrates that real-world AI integration requires honest assessment of current capabilities and limitations. Their investment in observability and custom agents suggests a maturing ecosystem where companies can extract genuine value from AI coding tools without overhyping their scope.
