Forward deployed engineers ship into client environments. You build the parts they ship. This role owns the internal agentic layer: orchestration, state and memory, tool interfaces, and the cost controls that keep an autonomous system from quietly spending a client's quarterly budget on a corrupted file.
It is a deep engineering seat with a short feedback loop. Every abstraction you write is in production somewhere within weeks, and you will hear directly from the engineer who had to debug it at midnight.
What you will do
- Design and maintain the orchestration layer: hierarchical control, delegation between specialist agents, and clean failure semantics
- Build the dual-memory architecture our systems rely on, spanning fast session state and long-term vector recall
- Own tool and integration interfaces, including MCP servers for the systems our clients run
- Implement cost guardrails, rate limiting, and the plan-and-approve flows that keep autonomy accountable
- Work with the forward deployed team to turn patterns that worked once into components that work every time
What we look for
- Excellent Python, with real experience designing APIs and abstractions other engineers depend on
- Deep hands-on work with LLM orchestration, whether on a framework or hand-rolled
- Practical knowledge of vector stores, retrieval quality, and where retrieval genuinely stops helping
- Distributed systems instincts: idempotency, retries, backpressure, and partial failure
- Strong opinions about reliability, loosely held
Nice to have
- Open source contributions to agent or LLM tooling
- You have built and operated an MCP server against a real production system
- Experience with multi-tenant systems and per-client isolation
Your first six months
- Month 1: ship your first component into a live engagement
- Month 3: own a major subsystem such as orchestration or memory
- Month 6: the architecture of our agentic layer reflects decisions you made and can defend