Position:Agentic Engineer - Multi-Agent Systems: Build, Connect & Ship About the Role:We're looking for an Agentic Engineer whose core craft is building agents - lots of them, fast. Not one chatbot over a document store: production agent systems that call enterprise tools, hand work to each other, and run where real users depend on them. You'll move quickly, iterate often, and own your work end to end.
Your job is to build on that platform and be fluent in its language: your agents register a manifest, call models and tools through gateways, and emit standard telemetry. The platform makes the plumbing disappear so you can ship agents at a pace that would be impossible if every project solved auth, deployment, and monitoring from scratch.
The fast lane agentic project team mission is to enable a digital and AI-enabled engineering workflow and organization. The fast lane team assignment is targeted to be 18 months as an Agentic-first project team, and then full-time deployment into the engineering organization.
This is an individual contributor role and the selected candidate will have the opportunity to work a hybrid schedule.
What We're Looking For You're naturally curious and self-directed - you build things because you want to understand how they work. You've containerized, self-hosted, and deployed real projects, and you've watched something behave differently in production than it did locally - and that made you a better engineer.
The clearest signal: you have agents that talk to each other and to tools, running somewhere that isn't your own machine - and ideally an MCP server that someone other than you depends on. You stay current with agent tooling by using it, not just reading about it.
Key ResponsibilitiesAs an
Agentic Engineer II, you will be responsible to:
- Design, build, and ship production AI agents - many of them - across frameworks (Claude Agent SDK, Microsoft Agent Framework, LangGraph, or custom harnesses)
- Build MCP servers that connect agents to enterprise systems (PLM, MES, ALM, data platform) - business logic only; the platform handles auth, deployment, and registration
- Design multi-agent workflows: orchestration, handoffs, and agent-to-agent communication patterns - deciding when one agent should be five, and when five should be one
- Build both the agent and the harness - the agent (model, instructions, tools, memory) and the harness it runs in (the loop: tool dispatch, context management, retries, state). Know when an off-the-shelf harness like the Claude Agent SDK loop is enough, and when the job calls for a custom one
- Run rapid experiments: hypothesis 12 prototype 12 deploy 12 evaluate 12 iterate - with evals that tell you whether an agent got better
- Embrace a fail-fast approach - short cycles, quick learning, and moving on with better information
- Evaluate agent frameworks, models, and tools through hands-on use, not benchmarks alone
- Document validated agent, skills and MCP patterns that contribute to the team's shared use case library - think in terms of scale: how do others reuse what you built
Technical Expectations- Minimum of 1+ years of relevant experience in the following:
- Shipped agents to production - plural, with real users,
- Agent SDKs & frameworks - deep hands-on with at least one of Claude Agent SDK / Claude Code, Microsoft Agent Framework, LangGraph, or equivalent - and willing to work across several
- MCP as a builder - you've written MCP servers, not just connected to someone else's tools
- Agent vs. harness fluency - you can explain the difference without hesitation, and you've built both: agents inside existing harnesses, and at least one custom harness of your own
- Python or JS/TS 12 LLM APIs 12 REST APIs & integrations
- Docker / containers and cloud deployment
- Evals - a working habit of measuring agent behavior before and after changes
- AI-assisted development (Claude Code, Cursor) as your default way of working
Strong Plus- RAG / vector search - useful as one component inside an agent, not the center of gravity
- OpenTelemetry / observability instrumentation for LLM apps
- Identity literacy - agent vs. user identity, OAuth on-behalf-of, scoped tokens
- Workflow automation (n8n / Power Automate)
- Local models (e.g. Ollama) 12 self-hosted infra
The Platform You'll Build OnYou'll work alongside the platform team and help develop it. That means the vocabulary of a governed agent platform should already make sense to you - or you should be excited to learn it in week one: model gateway, MCP gateway and catalog, agent manifest, entitlement vs. data authorization, telemetry traces, drift detection, kill switch.
Business Context- The 10%Strong technical skills are the foundation, but the ability to engage with business stakeholders matters too. You should be comfortable walking into a conversation with a process owner, understanding their actual problem, and communicating what you've built in plain language. Prior exposure to how business operations work - even in a non-technical capacity - is a genuine advantage.
Education Requirements- Masters degree in Computer Engineering, Computer Science, Electrical Engineering, Artificial Intelligence, Information Technology, Software Engineering, Data Science, Data Analytics, Information Systems, Information Science or Machine Learning required