Backend Engineer — Agent Experience

Deliveroo · London, United Kingdom - Deliveroo · 2-4 years experience · Full-time · Posted Oct 12, 2026

At a glance

Agent Experience builds the AI-powered backend services that power agent-assist tooling on our Astra support surface.
This role involves helping support agents resolve customer, merchant, and Dasher issues faster at the intersection of LLM-backed product features and high-volume, latency-sensitive support infrastructure.

Build and maintain backend services for agent-assist features on Astra, taking on feature-level work with increasing independence.
Translate requirements into working code: design the feature-level approach, write an RFC when the change affects other engineers, and implement it end-to-end.
Integrate with LLM-backed features — wiring up prompt/completion calls, handling provider responses, and validating output quality and correctness.
Break down assigned project work into tasks, provide accurate estimates, and keep your own progress and dependencies on track.
Apply AI tooling thoughtfully to your own workflow (coding, testing, review) and critically evaluate its output rather than taking it at face value.
Participate in code review — both giving and receiving — to help the team hold a high bar for correctness, quality, and standards compliance.
Collaborate with senior engineers and partner teams to clarify problem definitions, unblock yourself, and flag risks or blockers early.

Solid backend engineering fundamentals in Go, with experience shipping and maintaining production services.
Ability to independently own feature-level work: from a written requirement to tested, shipped code.
Comfort working in a domain with multiple chat/support surfaces — willingness to trace issues to the right service rather than guessing.
Clear technical writing (RFCs for feature-level decisions) and good judgment about when to ask for help versus push forward.
2–4 years of backend engineering experience, with a track record of growing ownership and scope.

Exposure to LLM-powered features in production (prompt/completion integration, provider APIs) is a plus — we'll ramp you up if you haven't worked with this directly.