The role:
We build and ship agent platforms that do real work in production. This role is for an engineer who designs self-running agent loops and multi-agent swarms, not someone who tweaks prompts one turn at a time. You will own systems that find work, do it, verify it, and report back, with humans at the gates rather than in every iteration. This is a hands-on, ship-first engineering role. You write production code other engineers trust, and you treat agents as software systems that need triggers, budgets, stop conditions, verification, and observability like any other.
What you'll do
- Design self-running loops. Define the trigger, scope, action, budget, stop condition, and reporting so an agent runs unattended, stays inside cost and iteration limits, and knows when it is done versus when to escalate.
- Build multi-agent swarms. Orchestrator plus specialized agents with clear file and task ownership, shared state or a shared mailbox, quality gates between stages, and handoffs that do not step on each other.
- Make verification first-class. Build the part of the system that can say no: the checks, evals, and reviewer agents that catch confident mistakes before they merge.
- Own agent state and memory. Persistent on-disk state and per-turn context assembly so long-running tasks survive restarts.
- Ship the platform around the agents. APIs, services, queues, and integrations in TypeScript and Node, deployed to AWS, with real tests, tracing, and observability.
- Keep humans in the loop where it counts. Plan approval and pull request review, and active management of comprehension debt.
What we're looking for
- Strong engineering fundamentals. 5+ years writing production software that other engineers depend on.
- Hands-on loop engineering. Experience designing agent loops with explicit stop conditions, budgets, retries, and self-verification.
- Multi-agent or swarm experience. Experience building or operating systems where multiple agents coordinate via orchestration, handoffs, and shared state.
- Fluency with modern agent tooling. Experience with Claude Code patterns, sub-agents, tool and function calling, MCP, and reason-act-observe loop patterns.
- Solid TypeScript and Node. Comfort with a service framework (e.g., NestJS) and a typed data layer (e.g., Prisma).
- Cloud and delivery. AWS (ECS/Fargate), Docker, and CI/CD proficiency.
- A verification mindset. Ability to build checks and evals that prove the work is correct.