AI Innovation Engineer

We are the leading online platform for comparing and buying business insurance. 

Not to toot our own horn, but we’re shaking things up in the SME business insurance market! We’re all about making it easy and efficient for small businesses to get the perfect coverage they need to stay protected. Say goodbye to headaches and hello to hassle-free insurance!

We're building an AI-first engineering team. We want an engineer who builds with agents and builds agents: someone who comes up with ideas, proves them fast with POCs, and ships the winners to production, safely, in a regulated insurance environment. You'll work within our guardrails, and when you see a better way, you're free to propose it and build it.

• →Build and prove: Build POCs quickly with LLMs, agents and MCP. Define success metrics up front and use evals to show quality, cost and business value before anything scales.
• →Engineer spec-first: Write requirements → design → tasks, hand execution to AI coding agents (Claude Code, Kiro) under human review, and keep specs, steering docs and CLAUDE.md in sync with the code.
• →Build reusable agent tooling: Package workflows as skills, subagents, hooks, plugins and MCP servers the whole team can reuse. Use hooks, not prompts, for rules that must always hold.
• →Orchestrate agents: Pick the right pattern (orchestrator/worker, pipeline, fan-out, routing), or one agent when one will do. Handle state, retries, token budgets and human approval gates, and run agents headless in CI/CD.
• →Ship and govern: Turn POCs into tested, observable services and RAG pipelines. Protect personal information, enforce permissions, keep audit trails and set agent autonomy boundaries.
• →Improve the framework: Spot bottlenecks and pitch stack changes backed by evidence.

• 3+ years in software engineering, including shipping LLM-powered applications.
• Agentic coding: Daily use of AI coding agents in real delivery work. You know when to use a skill vs a subagent vs a hook vs MCP.
• Spec-driven delivery: You've written specs an agent executed and reviewed its output critically.
• Agent orchestration: You've built at least one multi-step or multi-agent workflow and know how agents fail.
• Strong fundamentals: Python and TypeScript/Node, API design, testing, CI/CD and AWS (or another major cloud).
• RAG, evals and LLMOps: You've built a retrieval system and an eval harness, and instrumented tracing, cost and latency.
• Security mindset: You design against prompt injection, data leakage and excess agent permissions, and you're sceptical of complexity for its own sake.

You don't need experience with all of these, they're what we use today, and you'll learn them fast.