Senior Agent Harness Engineer

About Liaison

Liaison is an AI infrastructure and data company. Our platform connects to the systems a business already runs on, builds a unified view of its operations, and deploys agents that act on that context the way the owner would, bringing forward only the decisions that require a person. Our work spans agent engineering, model training outputs for small language models (SLMs), and frontier-level research, and we build our agent runtime and integrations in-house.

What month 12 looks like

The harness is measured: we can say how often agents succeed, where they fail and what each task costs. Model routing is tuned against those measurements, and computer use works reliably across tools that have no API.

What you will own

• The agent runtime and tool calling, including retries, failure handling and state
• Evals: how we measure whether an agent did the right thing, and regression checks on every change
• Model routing: which model handles which kind of call, balanced against cost and quality
• Computer use: agents operating software through its screen when there is no API
• Capture of agent actions, with the person's approval, edit or rejection, so we learn from real use
• Technical direction for the mid-level harness engineer

Required

• Has built and run an LLM agent system in production
• Has built an eval suite for an agent or model feature
• Has implemented tool calling with a real set of external tools
• Has owned a production system through incidents

Preferred

• Has worked on computer use or browser automation agents
• Has tuned model selection against cost and quality data
• Has led or mentored another engineer

Interview process

A 20-minute screening call, a paid work sample, then a final conversation with our CEO.