The posting
Founded in Switzerland in 1968, Zühlke is owned by its partners and located across Europe and Asia. We are a global transformation partner, with engineering and innovation in our DNA. We're trusted to help clients envision and build their businesses for the future – to run smarter today while adapting for tomorrow’s markets, customers, and communities. Our multidisciplinary teams specialise in tech strategy and business innovation, digital solutions and applications, and device and systems engineering. We excel in complex, regulated spaces including health and finance, connecting strategy, tech implementation, and operational services to help clients become more effective, resilient businesses.
If you share our values and want to do the best work, for the right reasons, we can offer you the chance to do it on a global scale and play a real role in shaping our exciting journey.
The Role
Zühlke is building its AI Factory: a production line for software in which AI agents turn an approved requirement into a verified change. People set the target, agents do the work, people assure the result. No agent approves, merges or deploys. Every quality gate produces evidence, and a named person decides.
As an Agentic Software Engineer you join the core team that builds this system, takes it into client projects and pushes it to production. You own the outcome: what you build is used by client teams and delivers the value it was built for. You design the agents, the workflows that connect them and the gates that check their output. That includes the harness the agents work in and the loops that run until a gate lets the result pass. You will also build agents and agentic workflows beyond software delivery, for clients and for Zühlke itself.
Tools and models change every few weeks in this field. We build on standard interfaces so that components can be swapped, and we expect you to enjoy that pace.
How You Will Make an Impact
- Design and build multi-agent workflows with clear hand-offs, durable state and explicit failure handling.
- Engineer agent loops that refine, plan, implement and review until an independent, default-fail evaluator lets the result pass. You define stop conditions, retry budgets and which model runs which step.
- Engineer the context agents work with: tools and knowledge access over code, documentation and domain, using MCP.
- Engineer agent harnesses: instruction files (AGENTS.md), skills, custom agents, hooks and CI checks, each concern in the right layer. Rules that must always hold go into tests and hooks, not into prompts.
- Implement spec-driven development with traceability from requirement to code and test.
- Build quality gates that run deterministic checks first, then evals and AI judgement, and end with a human decision backed by evidence. The checker is never the producer: a separate test agent writes hidden acceptance tests.
- Harden agents against prompt injection and run them in isolated sandboxes with least-privilege access.
- Make agent behaviour observable and auditable: an audit trail of agent decisions, cost per task and lead-time metrics. This trail is the evidence that governance rules such as the EU AI Act ask for, from human oversight to logging.
- Work directly with client teams to take agents, harnesses and loops into their projects and push them to production.
- Own the value after go-live: agree with the client what a capability should deliver, measure whether teams use it and whether it delivers, and close the gap when it does not.
What Is Important to Us
- At least seven years of professional software engineering, with a track record of owning code quality in production systems.
- You have built a working agentic system yourself, at work or in your free time, and use it to build software. In our selection process you demo it to the team, as a repository or live.
- Hands-on experience with agent orchestration: multi-agent workflows, hand-offs, state and failure handling.
- Experience with agent frameworks such as Claude Agent SDK, Azure AI Foundry or LangGraph.
- Experience with context engineering and the Model Context Protocol (MCP).
- Experience with evals as quality gates and with guardrails against prompt injection, and with tracing agent runs in tools such as Langfuse.
- Working knowledge of responsible AI and AI governance, for example the EU AI Act obligations on human oversight, logging and documentation.
- Harness engineering for AI coding agents such as Claude Code, GitHub Copilot or OpenAI Codex: you have written instruction files, skills, custom agents and hooks that a team relies on.
- Loop engineering: you have built agent loops with an independent evaluator, stop conditions and cost limits, and you can explain why a loop failed.
- Strong programming skills in Python or TypeScript; Java or .NET is a plus.
- Solid CI/CD and Git practice. Experience with GitLab, containers and Kubernetes is a plus.
- You write clean, tested, well-structured code. Prototypes that only work in a demo are not enough for us.
- You have taken software you built to production with clients and stayed accountable for whether it delivered the expected value.
- Delivery experience in regulated industries such as finance or health, where traceability is mandatory, is a strong plus.
- A team player who works closely with colleagues and client teams and explains technical trade-offs to non-engineers.
- You communicate confidently in English and work effectively in an international environment.
If you feel you don't meet all the requirements, we are still happy to get to know you, learn more about your ambitions and ideas and look forward to receiving your application!
We welcome people from all backgrounds, regardless of their gender, personality, national origin, race, religion, colour, sexual orientation, gender identity, age, marital status, disability or veteran status.
