Forward Deployed Engineer

Job description

Most AI roles keep you at arm’s length from the people the agents are supposed to help. This one puts you in the room with them. We’re building AI agents for risk managers, internal auditors, and compliance professionals, not assistants that surface suggestions, but agents that own complex, multi-step workflows end to end. Agents that an audit team can hand a task to and trust it will come back done. Getting from a promising agent prototype to something a Chief Audit Officer depends on is a completely different discipline to building the prototype. That’s what this role is about. As a Forward Deployed Engineer, roughly 80% of your time is with our biggest enterprise customers across Europe and the US. You’re embedded in their environments, sitting with internal audit teams as they plan an audit programme, with risk functions assessing their control landscape, with compliance teams navigating regulatory obligations. You’re not there to show them what the agents already do. You’re there to understand their world well enough to discover what the agents need to do next. You run discovery workshops, identify which workflows are genuinely ripe for agentic automation, prototype rapidly, test with real practitioners, and stay until the agent is reliable enough to be trusted. The other 20% is feeding everything you’ve learned back into the platform. The goal isn’t to accumulate bespoke agent builds, it’s to identify the patterns across risk and audit functions that should be generalised, so every subsequent engagement is faster and the agents get better for everyone. You sit in Product with a matrix into Engineering , which means you shape both the agent roadmap and the technical decisions that make agents work in production. This is not a sales role and it’s not consulting. You write code. You build evaluation infrastructure. You design golden datasets grounded in real audit and risk workflows. You catch agent regressions before a customer’s compliance team does. You’re accountable for outcomes. The Hard Parts of Agentic AI Are Your Domain Silent regressions when a model updates. Context window degradation across long audit workflows. Prompt instability under varied user inputs. Non-deterministic outputs in environments where a compliance officer needs to explain every decision to a regulator. You know these failure modes and you know how to build the guardrails, t…