ClickHouse Embraces Agentic Coding: From Boilerplate to Backend Development
Key Takeaways
- ▸ClickHouse has operationalized AI coding agents across multiple platforms and internal tools after strategic partnerships with Anthropic, Cursor, and Windsurf
- ▸Coding agents excel at boilerplate tasks and internal tool development but remain limited for complex backend C++ work
- ▸Claude Sonnet 4.5 represented a meaningful quality leap, enabling more sophisticated projects like the Team Productivity Dashboard (112-prompt session)
Summary
ClickHouse has shared insights into its practical use of AI coding agents after signing contracts with Anthropic, Cursor, and Windsurf. The company has moved beyond initial skepticism to deploy multiple internal coding agents (DWAINE, CAISER, TRAISA) alongside commercial tools, using them for boilerplate tasks, internal tools, performance tests, and dashboards. The organization acquired additional AI observability platforms Librechat and Langfuse to support this expansion.
While early versions of coding agents struggled with complex C++ codebases, Claude Sonnet 4.5 (released September 2025) represented a significant leap in capability. ClickHouse's team documented a complex project—the Team Productivity Dashboard—built through 112 prompts, demonstrating tangible value in agent-assisted development. However, the company acknowledges ongoing limitations with backend C++ development and emphasizes that coding agents remain most effective for specific use cases rather than autonomous development.
The company's measured approach reflects broader industry sentiment: coding agents are valuable tools for certain scenarios but not panaceas. ClickHouse's experience demonstrates that success with agentic coding requires clear understanding of appropriate use cases, proper integration with observability tools, and realistic expectations about current capabilities and limitations.
- Success requires proper observability and tooling integration, leading ClickHouse to acquire Librechat and Langfuse
- Practical adoption faces organizational challenges including employee uncertainty and the need to move beyond polarized views on AI's capabilities
Editorial Opinion
ClickHouse's pragmatic approach to agentic coding offers a refreshing counterpoint to the hype-versus-dismissal binary dominating AI discussions. By documenting real-world usage patterns, limitations, and incremental improvements, the company demonstrates that lasting value comes from understanding where tools excel rather than overselling universal applicability. The measured acknowledgment that agents work well for boilerplate but struggle with complex C++ codebases is honest and useful—exactly the kind of nuanced assessment the industry needs.
