Principal Agentic AI Engineer

Zoom•Seattle, WA

1d•$206,600 - $451,800•Hybrid

About The Position

ZoomMate (https://zoommate.com) seeks a Principal AI Agent Engineer, Research Engineer, or Applied Scientist to develop its AI Agent System framework and infrastructure. This position focuses on creating the agent harness, including runtime, orchestration, tool integration, memory, and evaluation layers. These components enable downstream ZoomMate Agent products. The role involves collaborating directly with product teams to implement the framework into practical applications. The successful candidate will contribute to delivering innovative agent experiences while ensuring seamless integration with various platforms and tools.

Requirements

5+ years of industry experience building production software and/or ML systems.
Practical experience developing agentic systems using LLMs, including tool use, planning, reasoning, memory, orchestration, or multi-agent coordination, implemented in production or research/prototype formats.
Programming fundamentals: data structures, algorithms, concurrency, and system design.
Excellent communication skills and a collaborative, ownership-driven mindset — you'll be building infrastructure other teams depend on.
Fluent in Python, and comfortable with at least one of Java/Go/TypeScript depending on the surface you're building.
Experience with agent frameworks and orchestration (LangGraph, LangChain, AutoGen, or an equivalent), RAG pipelines, or vector databases.
Experience with LLM post-training/fine-tuning, reinforcement learning, or building rigorous agent evaluation frameworks.

Responsibilities

Designing, building, and improving the ZoomMate agent harness, including orchestration, tool use, sandboxing, memory management, context handling, and multi-step reasoning loops for team collaboration.
Architecting the underlying agent framework and API/SDK surface, so Zoom product teams can embed ZoomMate Agent capabilities with minimal integration cost.
Building and integrating tool-use and interoperability layers, including MCP-style connectors and agent-to-agent (A2A) patterns, so agents can act across Zoom's ecosystem and third-party apps.
Designing and running evaluations and benchmarks for agent quality — reasoning, tool-use success, latency, and cost — and use them to drive iteration.
Owning the full lifecycle of the services you build — design docs, code review, deployment, monitoring, on-call — with production-grade reliability at Zoom scale.
Partnering closely with the product teams consuming ChatKit, turning their real integration pain points into better shared infrastructure rather than one-off fixes.
Integrating deeply with Zoom's engineering platforms, leveraging internal tooling to accelerate delivery.
Integrating AI coding agents like Claude Code and Codex into daily workflows, embedding that understanding into the design of tools tailored for such agents.