Applied AI Engineer

About the role

We're hiring two Applied AI Engineers to build the internal agentic tooling that runs OpenRouter — one focused on support, one focused on go-to-market. The two roles operate the same way; the difference is which domain you go deep on. We believe we can build software that fits our needs rather than buy software built for the masses. The best technical companies built proprietary tools, and you'll have the chance to do that at OpenRouter.

What You'll Do

• Build internal agentic workflows end-to-end on OpenRouter, from prototype to production, that automate support and GTM operations
• Integrate our systems with the tools the company runs on, and steadily replace those tools where it makes sense to
• Own the reliability of what you ship: evals, guardrails, and monitoring, so automation is trustworthy rather than flashy
• Drive measurable impact — deflect support volume, accelerate GTM motions, cut manual toil — and prove it with data
• Iterate on real usage: ship, measure, improve

What we're looking for

• Build-over-buy conviction. You build on internal tools first and champion them over polished external options, even when the internal path is slightly more painful. You buy only for genuine expertise, never for something buildable in a day. You push the last 20%.
• You've worked on internally-built tooling and made it good enough that people chose it. You're energized by reliability, edge cases, and polish, and you don't delegate the hard part.
• Domain fluency in support or GTM. Hands-on with ticketing, CRM, sales ops, or customer-facing workflows, so you know which work is actually worth automating.
• Architectural range. You understand systems end-to-end and can jump into any part of one.
• Deep backend orientation; front-end work is not the point here.
• A security mindset. You'll be building integrations with broad internal access. You think about permissions, blast radius, and the second-order effects of what an agent is about to do before it does it.
• Evals and data rigor. You measure automation impact quantitatively rather than asserting it.
• Excellence, time-boxed. A near-perfect bar, but disciplined enough to ship rather than polish forever.
• Nice to have: Experience across three or more agent harnesses. We don't need expertise in any one of them; we want to see that you've tried things. You've built your own agent harness for a workflowProduction LLM/agentic systems shipped with strong engineering fundamentalsOpen-source contributions to dev tooling or agent frameworks