$200,000 - $300,000
San Francisco, Hybrid
Permanent / Full-time
A product-led AI start-up is building one of the most widely adopted AI work companions in the world, operating at massive real-user scale with millions of daily interactions. The challenge has shifted to designing agent systems that can plan, reason, evaluate themselves, and operate reliably inside real products. This is an opportunity to work from first principles on agentic architectures that power production systems used by professionals globally.
Why This Role Matters
- Build agent systems that plan, act, reflect, and improve across complex, ambiguous user workflows
- Define foundational patterns for LLM tool-use, reasoning graphs, and self-evaluation in production
- Join at a point where agent architecture decisions will shape the long-term platform
- Work on problems beyond prompt engineering like runtime reliability, context limits, and learning flywheels
What You’ll Do
- Design and implement Plan–Act–Reflection style agent architectures
- Build DAG-based reasoning flows to deconstruct user intent into executable steps
- Develop agent skills including function calling, MCP-style integrations, and streaming APIs
- Solve runtime problems like context overflow / context rot through isolation, compression, and offloading strategies
- Architect automated evaluation and learning pipelines (reward functions, LLM-as-judge, RFT-style systems)
What You Bring
- Proven experience building and shipping agentic AI systems
- Strong understanding of workflow design, failure modes, and deterministic execution
- Comfort designing distributed systems, APIs, and protocols used across teams
- Practical experience with agent orchestration frameworks
