AI Agent Engineer

Gigi is an agentic operating system for enterprise media buying, with an AI media manager for Amazon DSP. The AI Agent Engineer will own the end-to-end development of agents and workflows, including APIs, data, tools, reasoning, execution, evaluations, and customer-facing product capabilities. The role also involves working directly with customers and building reliable AI systems for real advertising accounts.

Responsibilities:

• Ship agents end to end. You'll own capabilities from the data and tools an agent uses through its reasoning and execution to the interface a media buyer actually touches. Examples include turning a brief into a campaign, investigating performance, and carrying out approved changes. Our stack includes Python, FastAPI, LangGraph and LangChain; TypeScript, Vue and Nuxt; Java and Spring Boot; PostgreSQL, pgvector, Redis and NATS; and AWS
• Build with AI as a core part of your development workflow. Use Codex, Claude Code, Cursor, and other coding agents to help plan, implement, test, debug, and review software. You are still responsible for the architecture, quality, and correctness of what gets shipped
• Build the agent's judgment, not just its plumbing. Decide what Gigi should do autonomously and what it must escalate to a human, then score actions on risk and confidence so recommendations read differently when the data is thin
• Draw the line between probabilistic and deterministic work. LLMs for reasoning, structured software for execution. Knowing where that boundary sits on any given feature is a daily call, not a settled architecture
• Build reliable agentic systems. Design tools, retrieval, memory, orchestration, guardrails, and human approval flows that make Gigi dependable enough to take action inside real advertising accounts. Handle incomplete context, tool failures, and workflows that need to pause and resume. Prior experience with all of these is not expected
• Evaluate and improve agent behavior. Turn real customer workflows and production failures into evaluations and regression tests. Inspect traces, experiment with prompts and context, and measure whether changes improve the agent's ability to complete the task correctly
• Work directly with customers. You'll sit in on calls, read transcripts, and take feedback straight from the media buyers using the product. You'll turn ambiguous customer problems into shipped product without waiting for a perfect spec