Arbitr is moving beyond simple automation. We are building an autonomous enterprise — an army of intelligent agents running day-to-day business operations instead of people doing them by hand.
As an AI Ops Agent Engineer, you sit in the engine room. You find the manual processes eating the most internal time across Production, Sales, Marketing, Finance, HR and Customer Success, and you replace them with agents that work reliably in production.
This is a building role. You will architect, plan, build and integrate agents into Arbitr's existing agent suite. You will not be responsible for deploying or running the infrastructure — that sits with our IT & Infrastructure team — but you will need to understand how deployment works well enough to hand over clean, actionable requirements.
What You'll Actually Do
- Map processes. Sit with a function owner, understand what they do all day, and decompose it into steps a system can own. This is the part most people get wrong.
- Build automations and AI agents. Our core stack is Python (mainly Pydantic AI), plus n8n for lighter-weight workflow automation.
- Integrate, don't isolate. New agents plug into our existing suite and consume MCP tools built by our tool developers. Everything must compound into a product, not stack up as one-off projects.
- Write the evals. Create measurements of accuracy, cost and latency.
- Instrument everything. All agents are traced in Langfuse. Uncertainty is fine; silent failures are not.
- Design for failure. Define where the agent runs autonomously, where a human is required, and what happens when an external API goes down. Operations should degrade gracefully, never stop.
- Keep the boundary clean. Prompts, payloads and system instructions stay model-agnostic so we can swap providers without a rewrite.
What We Need
Mandatory:
- LLM fluency. Deep, practical command of how these models behave: prompt design, tool and function calling, structured outputs, context management, failure modes, and — critically — knowing when not to reach for a frontier model.
- Strong Python. You write code that other engineers can maintain.
- Process comprehension. You can absorb an unfamiliar business process quickly, spot the automatable core, and be honest about what shouldn't be automated. We weight this as heavily as the code.
- A bias toward shipping working things. We are under real pressure to remove manual hours. High-impact and unglamorous beats flashy every time.
Beneficial:
- Building eval frameworks and measuring agent quality systematically
- Autonomous, self-healing systems — retries, fallbacks, recovery without human intervention
- Deployment processes and CI/CD pipelines
- Model Context Protocol (MCP)
- Fine-tuning small or open-weight models for bounded, high-volume tasks
- Observability and tracing tooling (Langfuse or similar)
Working Here
You'll join a small, senior, multidisciplinary team split across Spain, Ireland and Auckland. The Auckland overlap is real: expect some early-morning calls, including a weekly team meeting. In exchange you get a genuinely fast environment with very little bureaucracy between an idea and production.
Not The Right Fit If
- You want to build top-level orchestration before there are agents to orchestrate.
- You'd rather research approaches than ship one that works.
Initially this will be a fully remote position - with a view to move to 3 days in Dublin office - with an employment contract based in Ireland.
Think this could be the right opportunity for you? We’d love to hear from you! If you’re excited about the role and feel your experience could be a great fit, don’t hesitate to apply. We can’t wait to meet you!
Originally posted on