Staff Engineer, Agents

LeanData helps the world’s fastest-growing companies automate, simplify, and accelerate revenue.

We are looking for a Staff Engineer to design and ship the production multi-agent systems at the core of LeanData’s new platform of autonomous agents for go-to-market teams.

This is a Staff-level individual-contributor role with founding-level ownership. You own the orchestration, tool-integration, memory, and coordination layers that let agents reason over go-to-market data and act reliably at enterprise scale — including the evaluations that prove they work and the path that safely writes their decisions back to a customer’s live data systems.

This role reports to the SVP of Engineering and is based in our Santa Clara, CA office. You are required to be in office Mondays and Wednesdays each week.

What you’ll be doing

• Build and ship production multi-agent systems end to end: agent control loops and planning, the orchestration graphs and tools they call, and the coordination, handoffs, state, and durability across agents
• Build the safe write-back path that applies agent decisions to a customer’s live systems without corrupting data
• Evaluate everything you ship: build the eval cases, rubrics, and regression tests that prove a change made the agent better
• Design agent memory and retrieval: persistent per-account context, pre-compute, and just-in-time lookups that keep reasoning fast, cheap, and reliable across many concurrent agent instances
• Harden against adversarial data so a crafted account name or note cannot hijack the agent
• Integrate frontier LLMs behind a model-agnostic abstraction with routing by task, cost, and latency; own the cost, latency, and reliability of your surface, and create the patterns that let the team build agents faster

Requirements

• 4+ years building production systems, including 2+ years shipping LLM-powered or agentic systems
• You have shipped customer-facing LLM or agent systems that real users depend on, and can explain how they failed and how you fixed them
• Strong, current Python (TypeScript or Go a plus)
• Hands-on with a modern agent framework / orchestration (e.g. LangGraph or agent SDKs), tool/function-calling, structured outputs, retrieval (RAG), and context/memory management
• Strong systems-engineering fundamentals (concurrency, distributed systems, statefulness, latency and cost at scale), plus skill debugging deep, non-deterministic failures in multi-step agent traces
• High agency: you scope, prioritize, and ship without waiting for permission
• Bonus points if you have
• Experience with Salesforce APIs (Bulk 2.0, Composite, Pub/Sub) or another large, messy enterprise data source
• MCP (Model Context Protocol), A2A, or similar tool and agent interoperability standards; the modern eval/observability stack (Promptfoo, Braintrust, Langfuse) and durable execution (Inngest, Temporal)
• Run hundreds or thousands of concurrent agent instances (serverless or function-style runtimes); multi-tenant data isolation (Postgres + RLS) and a strong security posture for holding customer data
• A founder or founding-engineer background, or contributions to open-source agent or LLM tooling