About the role
You build the part of Private Agent One that turns a person's intention into completed, inspectable work. Inspectable is the load-bearing word: the person has to be able to see what their agent did and why, which shapes the runtime more than any performance requirement does.
The work
Implement durable execution, tool calls, background jobs, checkpoints and human approvals. Keep one authoritative identity and policy domain per user while allowing constrained workers. Enforce permissions outside the model and make cancellation, revocation and uncertain external outcomes visible.
What good looks like
In your first 90 days, ship a useful workflow that survives interruption, requests approval at the right boundary and explains the actions it completed.
Evidence we look for
Bring strong backend or systems engineering and practical agent experience. Show attention to retries, secret handling and external side effects.
What we need to see
- Strong backend or systems engineering, plus practical agent work
- Careful handling of retries, secrets, and external side effects, with examples
- You design so an action can be explained after the fact, not just executed
- You treat an irreversible side effect as something to be deliberate about
Nice to have
- Workflow or orchestration engine experience
- Idempotency and exactly-once semantics in practice
- You have built an audit trail somebody actually used
The exercise
Design a workflow where the user revokes permission while an external task is still running, including the limits of recall.
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.