Staff Engineer — Agentic AI

### About the Role This is a senior technical leadership role at a well-funded Series A startup building AI agents for hardware engineers. You'll own the core agent intelligence layer — the system that translates mechanical engineers' intent into reliable, cost-efficient multi-step workflows across complex desktop engineering tools (CAD, simulation, PLM, and more). You'll report directly to the CTO and serve as technical lead for a small team of AI engineers, a user researcher, and domain expert contractors. This role sits at the intersection of applied agentic AI, user research, and product delivery — and it directly determines the product's real-world value to enterprise customers. The company serves Fortune 100 customers and has significant backing from top-tier investors. This is a high-impact, on-site role based in San Francisco. ### What You'll Do - **Drive agent task success rate.** Own the metric that matters most — define the eval framework, establish baselines, and systematically improve whether the agent can complete the workflows engineers actually need. - **Set and enforce token budgets per problem.** Define per-task token budgets, track cost per completed workflow, and ensure commercial viability — not just technical impressiveness. - **Build rigorous evaluation infrastructure.** Design benchmarks grounded in real user stories with SWE-bench-level rigor — reproducible, adversarial, and tied to measurable customer value. - **Lead user story mapping and validation.** Work directly with the user researcher and domain experts to interview engineers, document workflows in detail, and validate that what you're building against reflects reality. - **Expand workflow coverage.** Systematically grow the percentage of top user story steps the agent handles end-to-end, prioritizing by customer value and technical feasibility. - **Translate user stories into evals.** Close the loop between user research and agent benchmarking — every validated user story becomes a test case. - **Own the agent architecture.** Make foundational decisions on tool-calling strategies, state management, error recovery, model routing, and context management. - **Lead as a player-coach.** Set technical direction, review architecture decisions, write production code, unblock the team, and raise the engineering bar. - **Collaborate cross-functionally** with integrations, product, and customers during POCs to align agent behavior with real-world enterprise usage. ### What We're Looking For **Dealbreakers (must-haves):** - 7+ years in software engineering, with at least 2 years building agentic LLM-based systems — agents that call tools, manage multi-step workflows, handle failures, and operate under cost constraints. - Deep experience with LLM application architecture: model selection, context window management, retrieval strategies, tool-calling frameworks, and orchestration patterns. - Strong evaluation and benchmarking instincts for agentic systems — task completion, cost efficiency, failure mode analysis; familiarity with benchmarks such as SWE-bench, GAIA, or τ-bench. **Required:** - Proven track record of shipping AI systems with measurable outcomes (agent task success rate, cost efficiency) — not just demos. - Strong Python skills and working knowledge of the LLM tooling ecosystem: function calling, tool use APIs, tracing/observability tools (e.g., Logfire, LangSmith), and evaluation frameworks. - Experience leading a small technical team (3–6 engineers): setting technical direction, performing code reviews, driving architecture decisions. **Nice to Have:** - Published work or open-source contributions in agentic AI systems. - Familiarity with enterprise deployment constraints — agent behavior on locked-down corporate workstations. - Experience with desktop automation, COM, or programmatic control of applications (beyond web APIs). - Background in mechanical engineering, CAD/CAE, PLM, or adjacent industries. - Experience building or contributing…