AI Engineer, Staff - AI Harness

Staff Engineer, AI Harness Remote-first (Europe) · London hub The company An early-stage, well-funded startup building a first-principles computational model of raw material supply chains; the polymers, chemicals, metals and minerals behind every physical product. Procurement decisions worth billions are still made on averages, assumptions and supplier claims. We model the physics instead, so those decisions can be computed. Small, flat team. No middle management. Every person owns outcomes end to end. The role Every company has access to the same frontier models. What decides whether a professional will stake real money on an answer is everything around the model — the tools, the context, the verification. That layer is the harness, and it's where the engineering lives. A procurement lead asks whether a supplier's surcharge is justified, or what happens to their polymer cost if crude hits $X. The system decomposes that into tool calls against the domain model and physics computation, and returns an answer that is computed, traceable and defensible in a negotiation. You own the design, the plan and the build. What you'll do * Design the tool layer that exposes the physics model as composable tools agents can orchestrate, and set the standards other engineers build against * Build the agent orchestration, context and memory, grounding and provenance, and the runtime guardrails that make it fail safe * Ship the eval machinery — every module ships through evals, correctness is measured rather than asserted, wrong answers can't ship silently * Build the benchmark that shows, module by module, how much the full harness beats a general-purpose model on real procurement work — then drive our modules past it, one at a time * Get live agents running a customer's multi-step procurement workflow end to end, grounded in their own data * Work full-stack when the answer needs to surface in the product Later: the harness moves from answering to acting, and where a specialised model beats orchestrating a frontier one on cost or accuracy, you lead us into training our own. What we're looking for * You've designed and shipped production platforms that enterprise clients depended on, and can walk us through one you built and why it worked * You've built a production AI or agent system that ran reliably — where the logic lived in the system rather than the model, context was engineered and budgeted, correctness was measured with real evals, and cost, latency and failure modes were designed for rather than discovered * Expert Python, comfortable enough in TypeScript/React to land results in a real UI * You've owned a technical domain end to end and raised a team's ceiling through the systems and tooling you built * You direct agents in your own workflow and would rather build the tool than repeat the task Useful but not required: durable orchestration or long-running workflow systems, serious tool-use and function-calling work, retrieval beyond tutorials, applied AI where wrong answers have consequences, supply chain or manufacturing exposure. Stack Python FastAPI, Neon Postgres, Dagster, AWS. LangGraph, MCP, Pydantic AI. Next.js/TypeScript. AI-native development workflow throughout. Experience with the exact stack is a signal, never a requirement. If you are Staff level with the right skills and passion for AI, please apply and we will be in touch.