About the role
One person's agent should work across many machines without ever losing track of who is in charge. You make delegation and recovery predictable, which is a distributed-systems problem with an unusual constraint: the authority boundary is a single human being, and it must hold even when parts of the system are lying or gone.
The work
Build durable task state, queues, leases, fencing, cancellation and reconciliation. Design failover so a recovered replica cannot become a second conflicting authority. Distinguish control-plane replication from execution workers and expose uncertainty when external outcomes cannot be confirmed.
What good looks like
In your first 90 days, demonstrate delegated work across two devices, interruption recovery and prevention of duplicate externally visible actions in the supported test cases.
Evidence we look for
Bring strong reasoning about consistency, retries, failure detection and distributed transactions. Show a system you operated through real failures and the guarantees it actually provided.
What we need to see
- Strong reasoning about consistency, retries, failure detection, and distributed transactions
- You have operated a system through real failures and can say what guarantees it actually provided, as opposed to claimed
- You design for partial failure as the normal case
- You can explain a consistency model to somebody who will have to implement against it
Nice to have
- Consensus protocols in production
- Formal methods or model checking
- Edge or intermittently connected systems
The exercise
Trace a task whose worker finishes after its lease expires and whose external API response is lost.
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.