Design and implement agentic features end-to-end, from prototype to production.
Build and maintain agent loops, tool layers, context management, persistence, sub-agent orchestration, observability, and safety hooks.
Define quality standards for agent performance by creating golden datasets, writing evaluations, and setting up regression checks.
Instrument agents with proper tracing and telemetry, and investigate production behavior using real execution traces.
Work directly with client stakeholders to elicit and refine requirements, translate them into technical scope, and align on expected outcomes.
Make pragmatic build-versus-import decisions when selecting frameworks, tools, and external components.
Participate in architecture discussions and code reviews, contributing to scalable and production-ready engineering decisions.
Move between projects and priorities as needed, taking ownership of delivery in a fast-changing environment with a high degree of autonomy.

2–3+ years of commercial engineering experience
Ability to build a tool-using agent loop from scratch, including tool calling, structured outputs, stop reasons, parallel tool use, and error recovery
Strong understanding of context engineering, including what to write, select, compress, and isolate
Hands-on experience with modern agent frameworks or SDKs such as LangGraph, Claude Agent SDK, OpenAI Agents SDK, or Pydantic AI
Working knowledge of Model Context Protocol (MCP)
Experience building or extending evaluation harnesses, solid Python and/or TypeScript skills, and Upper-Intermediate English

Middle AI Agent Engineer

Key skills