AI Product Engineer, ClickHouse

• You've been building agents long enough to have opinions — about context engineering, tool design, when to use a skill vs. a tool, what evals catch and miss, and where popular frameworks break down
• You think in production terms: p99 latency, cost per task, whether the system still works next week without intervention
• You move quickly, ship often, and learn from what breaks
• You care about developer tools and have a clear sense of what good DX looks like
• You do well with ambiguity and ownership
• 5+ years of software engineering experience, including 1–2 years on LLM-powered systems or agents in production
• Strong backend skills in TypeScript/Node.js and/or Python. Comfortable in both, even if one is primary
• Hands-on experience building agents: multi-step tool use, planning, memory, error recovery. You've shipped them and dealt with the failure modes
• Experience designing skills (Markdown-based workflow encodings, Anthropic-style or similar) and a clear view on when a skill, a tool, or both is the right fit
• Experience with MCP: building servers, designing tools, and thinking through auth, scoping, and observability for agentic systems
• Strong evals practice: golden sets, LLM-as-judge, regression detection
• SQL proficiency — you can write ClickHouse queries directly
• Comfort with Docker and Kubernetes
• Active in open source and the developer community

Desirable

• We're hiring an AI Product Engineer to build agentic capabilities on top of a petabyte-scale observability platform, with a focus on developer experience
• If you've been building agents, designing skills, and wiring up MCP servers — and you've hit the limits of generic copilots for production work — we'd like to talk
• Build agents that investigate incidents. They surface anomalies, answer "why is production broken?", and use ClickStack as their substrate
• Write skills, not just prompts. Build a library of reusable skills that captures how our team debugs, finds root causes, writes ClickHouse queries, and runs incident response, so agents pick up the right playbook instead of starting from scratch
• Own the agent stack end-to-end. Context engineering, tool design, evals, tracing, cost. You're responsible for whether the agent works in production
• Make ClickStack a great place to run AI workloads. Build the MCP servers, SDKs, and integrations that let customers' agents read telemetry, take action, and stay observable themselves
• Work in the open. Collaborate with OSS contributors and customers, debug their problems alongside them, and feed what you learn back into the product
• Tackle the hard parts. Latency, cost, context window limits, eval coverage, hallucinations on real telemetry