You will own the agentic infrastructure at a seed-stage AI startup building the foundational layer for AI-native service operations. This means designing and shipping the systems that let agents execute reliably in production, retain months of client context, and know when to hand off to a human. This is a foundational engineering role at a very early team, with outsized scope and impact from day one.
What You’ll Do
•
Build and maintain core infrastructure that enables agents to execute tasks reliably across dozens of iterations.
•
Design and implement memory systems that retain months of client context from messy, real-world operational data.
•
Develop eval harnesses that teams actually trust to make production shipping decisions.
•
Own the full loop: build it, measure it, break it, fix it, and make it learn.
•
Ship agent systems for real customers and iterate directly based on production feedback.
•
Contribute across infrastructure, orchestration, customer collaboration, and early hiring as the team grows.
What We’re Looking For
•
3+ years shipping LLM agents that ran unattended in production for real users, with a clear understanding that the model is the easy part.
•
Hands-on experience building eval harnesses that were actually used to make production deployment decisions.
•
Experience with retrieval and memory systems that work on real, messy operational data, not demo RAG pipelines.
•
Strong Python proficiency for building production systems end to end.
•
Experience with agent orchestration, workflow management, or reliability patterns such as failure recovery and retry logic.
•
Experience building agent tooling, frameworks, or libraries is a strong plus.
•
Experience with human-in-the-loop or approval workflow systems is a plus.
•
Track record of intensity and execution: you force things into existence; open-source contributions or a founded technical project are a bonus.