AI Engineer
Own AI systems end to end, from the first client conversation to production. Evals, governance, MCP, and code the client keeps.
38Shift builds AI systems companies can actually run. We are an Anthropic Claude Network Partner and a member of the OpenAI Partner Network, working with clients across Europe.
Most AI projects die in the same place: the gap between a demo that impresses and a system a team trusts on a bad day. That gap is our business. One scoped problem, in production in 30 to 60 days, with evals, monitoring and escalation paths. The client owns every line. If they fired us tomorrow, it keeps running.
We build ahead of the curve, not behind it. We shipped WebMCP on our own site before most companies knew the API existed. We run our own operations on the agents we sell. We treat governance as a product decision: what the model can reach, when it acts, when it defers, who reviews it.
Small team, senior seat. You own systems end to end and make the architecture calls yourself.
- Claude API and Claude Code
- OpenAI
- MCP
- RAG and vector search
- Custom eval harnesses
- n8n
- TypeScript / Node.js
- Python
- Postgres / Supabase
- Next.js / React
- Vercel / Hetzner / Docker
What you will do
- Take one scoped client problem to production in 30 to 60 days, and stay with it when it breaks.
- Write the eval set before the build. Real inputs, run in CI, so a model or prompt change is measured instead of guessed at.
- Build the reliability layer: structured outputs, retries, fallbacks, confidence thresholds, escalation to a human where the risk is real.
- Make the calls. Which model per task, where retrieval grounds the answer, n8n or code, one agent or two.
- Ship the governance layer: access boundaries, tenant isolation, audit trails, review checkpoints. In health and finance work, this is the product.
- Build MCP servers and tool schemas that other people's agents can use.
- Hand it back clean, documented well enough that the client's engineers extend it without you.
- Sit in discovery and say what should be an agent, what should be a script, and what should stay human.
Who we are looking for
- 3+ years shipping real software, backend depth in TypeScript/Node.js or Python.
- Production LLM experience: prompting, context engineering, agent architectures, retrieval, evals. You have shipped something that had to keep working after launch.
- Empty repo to deployed system without waiting for a spec.
- Comfortable in front of clients. You can hold a technical position without hiding behind jargon.
- You keep up with the field because you find it interesting, not because someone assigned it.
- Strong English.
Nice to have: regulated environments, MCP, n8n, retrieval at scale, Next.js.
Why 38Shift
- Your name on systems clients run their business on.
- Direct line to the founder and to clients. No layers, no ticket queue.
- Partner access to frontier models and the roadmaps behind them.
- Fully remote, flexible hours.
- Claude Max for everyone, generous hardware budget.
- Room to grow fast, because the company grows when you do.
Sound like your kind of problem?
Send your CV to careers@38shift.com. Tell us what broke in production and what you did about it.