We are building the agentic foundation beneath Maersk’s data and AI platform, allowing people and their agents to safely build and operate production systems at scale. Our ambition is to enable a small team, working effectively with its own agent fleet, to deliver what previously required an organization many times its size. 

This is not a role for someone who uses AI as autocomplete. We want engineers who have fundamentally changed how they work. 

What “agentic” means here 

You work with multiple or long-running agents as an extension of your engineering capability. You turn ambiguous problems into parallel, bounded workstreams, provide context, review decisions and verify what agents produce. 

You treat idle agent capacity as a problem. Agents should be investigating production issues, building regression tests, benchmarking alternatives, mapping unfamiliar repositories or testing ideas that could remove weeks of future work. 

This is not manufactured busywork. It is a prioritized backlog that continuously produces evidence, code, tests or learning. 

An underdefined task is not an invitation to transfer ownership. You identify constraints, compare alternatives and form a point of view. You make reversible decisions within your ownership and bring evidence-backed recommendations for consequential ones. 

You are comfortable where documentation and established patterns stop. You map unfamiliar systems, form hypotheses, run experiments and bring back evidence — not just ideas. 

What you will build 

You will create a platform where engineers can build, deploy and operate agentic systems without repeatedly solving the same foundational problems. 

This includes: 

• Agent runtimes, orchestration patterns, golden paths, SDKs and self-service capabilities. 

• MCP servers, tool catalogues and governed access to enterprise systems. 

• Agent identity, delegated actions, permissions, budgets and audit trails. 

• Evaluation gates, shadow deployments, regression suites and safe rollouts. 

• Traces, execution histories, retries, backpressure, quotas and cost controls. 

• Memory, RAG, caching, model routing, local inference and fine-tuning where appropriate. 

• Vendor-independent seams that keep business logic portable across models and frameworks. 

Our domain teams use technologies such as LangGraph, Temporal, Dify, AWS, Foundry, Databricks, ADK and open-source components. We are loyal to none of them. Your job is to identify durable patterns and turn them into reusable capabilities. 

Your responsibilities 

• Own platform capabilities from design through production, including reliability, security, performance and cost. 

• Use your agent fleet to explore repositories, implement solutions, build tests and investigate failures. 

• Verify agent output through code review, tests, evaluations, traces and benchmarks, and improve workflows when output fails. 

• Work with domain teams to understand their needs and turn recurring challenges into reusable tools and platform capabilities. 

• Contribute to technical direction by evaluating alternatives, explaining trade-offs and making decisions within your area of ownership. 

• Support less experienced engineers through mentoring, pairing and constructive code reviews, helping them build confidence and take ownership. 

• Share engineering practices across the team, including how to scope work for agents, challenge their output and verify results. 

What you bring 

We want people who have shipped agentic systems used by real users, can navigate unfamiliar systems and understand production engineering. Experience with MCP, orchestration, evaluations, Kubernetes, infrastructure as code, observability, distributed systems, identity or authorization is valuable. 

We welcome different levels of experience and do not expect one person to match everything. The scope of ownership and technical leadership will reflect your experience and strengths. We care more about what you can build, operate and prove than your current title. 

Open source and a fast-moving field 

Open source is part of our product surface. When a dependency lacks an important capability, the default should not be another permanent internal workaround. Where practical, you contribute the implementation, tests and documentation upstream. 

AI changes weekly. You test relevant developments against real workloads and recommend whether to adopt, watch or discard them. 

While applying what to consider? 

Send us your CV or profile, together with evidence of: 

• An agentic system you shipped and how agents fit into your workflow. 

• An ambiguous problem or incorrect agent output you investigated and resolved. 

• An AI development or open-source improvement you tested or contributed to. 

• Links are more useful than adjectives. If this changes your pulse, we should talk. 

#LI-SS1