About the role
You build the product people actually touch, across the whole stack, on a team small enough that the thing you ship on Tuesday is in front of users by Friday. This is AI-first work in the literal sense: the agent runtime, the data underneath it, and the interface on top are one problem, and you will work on all three rather than being handed a slice.
What you’ll do
- Build Agent One features end to end - from data and agent runtime to a UX that feels inevitable
- Make AI a first-class material: tools, memory, evals, and the consent layer underneath
- Design data systems that are fast, private, and correct - receipts, least privilege, minimal disclosure
- Obsess over latency, accuracy, and control; instrument outcomes so we don't fool ourselves
- Write small, modular, legible code others could lift on its own
What we need to see
- You have shipped and owned production full-stack work, front to back, for real users
- Hands-on LLM and agent work: tool use, context handling, and the evals that keep it honest
- Taste in interface, not just capability: you can tell when something is technically correct and still bad
- You debug across boundaries rather than escalating at the edge of your usual layer
Nice to have
- On-device or edge inference experience, since much of our compute is on hardware the user owns
- You have worked somewhere with a real privacy or consent constraint, not just a policy page
- Open-source work anyone can read
What winning looks like
- Features shipped that move activation and weekly-active retention
- Latency, accuracy, and quality budgets met
- Code others reuse; low defect and regression rate
Where and how we work
In the office together five days a week, in any of these cities. Remote-friendly around your family, arranged one person at a time.