Lead AI Engineer - 11510

What You'll Do

• Build and extend the eval pipeline: how we measure task completion, catch regressions, and turn production failures into fixed behaviors.
• Work on context management: what the agent sees, when, and how we keep long-horizon tasks coherent without burning tokens.
• Design sub-agent patterns: when to fan out, how to compose specialized agents cleanly, and how to keep the parent agent in control of the outcome.
• Own document parsing: turning supplier-uploaded invoices, catalogs, and contracts into structured context the agent can reason over.
• Wrap Coupa supplier APIs as agent-callable CLI tools, with clean error surfaces and sensible defaults for an agent caller.
• Ship supplier-facing skills on top of the harness: the procedural instructions and tool compositions that let the agent handle specific tasks end-to-end.
• Debug messy production behavior: why did the agent take that path, where did it get confused, and what tool, context, or harness change fixes it?
• Partner with the senior engineer on harness architecture calls as you find gaps while working across the stack.

What You Will Bring to Coupa

• Have 3–5 years shipping production software and are strong in Python (TypeScript a plus).
• Have shipped at least one LLM-powered feature or product, even if small, and can talk about what went wrong and what you'd do differently.
• Are a daily, heavy user of agentic coding tools — Claude Code, Cursor, Codex, or equivalents.
• Have side projects. Real ones. Things you've built because you couldn't stop thinking about them. Ideally things the new generation of AI tools made possible for you to finish.
• Write and speak English fluently. Skill authoring is prose-heavy and the team operates in English.
• Debug by reading, not by guessing. You reach for logs, traces, and evals before theorizing.