Software Engineer, Agentic AI Systems

We get a lot of applications from people who've built LLM wrappers, written a few prompts, and called it AI engineering. That's not this job.

We build multi-agent systems that run in production for enterprises and scale-ups across the world. Our agents call tools, hold state across long horizons, recover from bad tool outputs, and stay inside budget caps the finance team actually checks. When something goes wrong at 2am on a Friday in London, someone on our team has to know why.

**The stack**

Fundamentals beat framework familiarity. An engineer who understands distributed systems, state management, and good API design will pick up our tools in weeks.

Day to day: GCP is our primary cloud, with AWS for specific client deployments. Agents on Google ADK with MCP for tool integration. Frontends in Next.js and TypeScript. Everything containerised, with Terraform and proper CI/CD.

If you've built serious systems on AWS and not GCP, we're interested. If you've shipped LangGraph or AutoGen and not ADK, we're interested. Reasoning about the problem beats memorising an SDK.

**What you'll own**

Multi-agent workflows end to end: the agent graph, the tools, the eval harness built before a client asks for one, the instrumentation that lets you actually debug it, and standing behind the output when a senior client pokes at it.

You'll write backends that don't fall over. Frontends that don't embarrass the rest of the work. You'll sit in client calls and explain the system in language a CTO and a CFO can both follow.

You'll push back on us when we're wrong.

**What we actually need**

Years of experience isn't the filter. The shape of your thinking is.

_Fundamentals._ You write code your teammates want to read. You can keep a system debuggable at ten times the load it was built for. You've been wrong in production and learned the right lessons.

_Cloud._ Serious hands-on GCP, AWS, or both, with a current professional-level cert. GCP is where our work lives, so if you're AWS-only, tell us how you'll ramp.

_AI systems._ You've shipped something agentic that stayed up. You've debugged an agent loop burning money, losing state, or silently retrying itself into oblivion. Your view on evals goes beyond "the outputs look okay."

_Communication._ You turn messy client briefs into shippable architecture without three meetings. You write technical docs an executive can decide from. You disagree with a co-founder in a way that makes the decision better.

**The unicorn profile**

Most of what makes this job hard isn't the stack, it's the combination. Clean backend code, a real opinion on retrieval, a frontend that doesn't look like a hackathon project, and client emails that don't need editing. You know when multi-agent is right and when one agent with better tools is.

You've shipped evals, not just written them. You reason about token budgets the way a backend engineer reasons about query plans. If you've contributed to an open source AI framework, written content other engineers cite, or been the person who got paged when the agent went sideways, we want to talk.

**What we offer**

Genuinely hard work on problems that matter to clients paying real money. A small senior team that treats your judgment as load-bearing from day one. Direct access to both co-founders. GCC and APAC exposure without relocating. No middle management. No tickets. No long standups.