Senior AI Native Engineering

We're looking for a Senior AI-Native Engineer to join the AI-native team. This role is about designing AI systems that other engineers and customers depend on, taking them to production, and measuring whether they work. The role works the AI-native way: spec first, tests first, coding agents doing implementation under review. Responsibilities Design and run production LLM systems: multi-stage pipelines or agents with a curated tool set, output schemas, validation, retries and failure recovery Build and maintain evaluation: golden datasets, regression evals on every prompt change, LLM-as-judge with spot checks, production telemetry Decide where work is autonomous, where it assists, and where a person approves Deliver spec-first and test-first with coding agents, and keep project context files (AGENTS.md, CLAUDE.md, steering files) deliberate and token-aware Put AI steps into CI/CD (review gates, test generation, quality hooks) and tune them against false positives Estimate AI cost before building, measure it after, and use the cost levers (model routing, context trimming, caching, batching, limits) Publish reusable skills, prompts, tool schemas, eval sets and MCP servers, and keep them current Raise AI failures early and fix them structurally. Help new teammates adopt the team's AI workflow Requirements At least one AI workflow or system running in production or CI without daily attention, which the candidate can explain: trigger, users, own part, and how they know it works Multi-step LLM or agent design experience: tool restriction, output schemas, validation, fallbacks, latency trade-offs An evaluation approach the candidate built or extended A wrong output fixed with a structural change, not a re-run Knows what drives the cost of their own AI work and has used at least one cost lever with a number on the effect 5+ years of professional software engineering with production ownership 2+ years building with LLMs or agents on real projects Clear communication: a peer could rebuild their work from their description Nice to have Multi-agent systems in production Brownfield discovery on a legacy codebase AI steps in CI/CD with measured outcomes (PR cycle time, defect density, AI acceptance rate) More than one model provider or platform with reasons for each Tracing and observability for LLM systems Experience in security-sensitive environments