Sr. AI Engineer – Agentic Systems

The Role

As a Senior AI Engineer on agentic systems you'll own the agent architectures that put our models in front of business users, and the work of making them reliable enough to stay there. Design, agentic behaviour, and performance optimization are all inherent to the role, because the needs converge: architecture shapes behaviour, and behaviour determines what needs optimizing.

What makes it interesting is what the agents have to do. Underwriting, claims, and the other core areas each involve multi-step workflows with real consequences, so agents need to reason over long context, follow complex instructions reliably, and integrate with systems of record that were never designed for them. Reliability is the hard part, and it is largely an evaluation problem rather than a prompting one.

This work is aimed at direct implementation. People on this team move between agentic work, systems engineering, and applied science as priorities shift. Less a multi-agent system than one generalist with broad tool access.

Major Responsibilities

Design and develop agentic systems end-to-end: agent loop design, orchestration, memory and state, tool integrations, and multi-agent workflows

Make agents reliable: diagnose why a loop stalls, why a tool gets misused, or why behaviour drifts between runs, and design the guardrails that hold up under real traffic

Measure agent quality continuously rather than at demonstration time: evaluation sets that reflect real workflows, regression checks that catch behavioural drift, and analysis that explains a failure rather than only flagging it

Tune the tradeoffs that decide whether an agent is usable: latency, answer quality, and token cost, including the judgement of which to give up in a given workflow

Integrate agents with the systems and data they depend on, and carry them from prototype to something the business can rely on in production

Qualifications
What You'll Bring

Deep expertise in production Python, with strong working knowledge of current agent frameworks and protocols (Claude Agent SDK, deepagents, LangGraph, MCP, or equivalents)

Experience building and debugging agentic systems in production, where failure modes are emergent rather than exceptions: loop instability, tool misuse, unbounded context growth, and non-determinism

A feel for the performance envelope: where latency comes from in a multi-step agent, what a token budget buys, and how to trade quality against both

Evaluation rigour, and the judgement to design measurement that holds up in a business environment rather than a demonstration

Fluency with agentic coding tools in your own workflow, such as Claude Code and Codex. The team is fully immersed in this way of engineering, and we expect it to be part of how you build rather than something reached for occasionally

The profile can come from either direction: a software engineer who has moved into AI systems, or an AI engineer with strong systems fundamentals. Either way we expect engineering discipline, meaning version control, tests, reproducibility, and code the next person can pick up

Strong Preference Given To

Evaluation and tracing tooling for agent behaviour (LangSmith, Weights & Biases, or custom solutions)

Distributed and event-driven systems experience (Kafka or similar)

Retrieval and memory system design for long-running agents

Full-stack development experience integrating agents into user-facing applications