AI Engineer

The prompt is the easy part.

Maybe your situation looks like this.

You got a Claude Code or Copilot license at work this year. Everyone did. But finishing your tasks faster isn’t enough for you. You want more. You compare harnesses. You try a new model the day it comes out. You build a loop of agents, then a graph of them. You put a workflow behind an endpoint, try a vector database just to see how it behaves, build a semantic index over your own notes, ship a small app nobody asked for. You want this to be your real job, but you still have to take the next ticket.

Or maybe like this.

You’re past all that. You have the Max accounts and the API keys. Your AI bill runs to tens of thousands of kronor a month and you think it’s cheap. You build the foundations, push what an autonomous model can actually do, and you’re good at it. The problem is that you’re alone. Nobody at work does what you do, and when you figure something out, there’s nobody to argue with about it.

If either of these is you: read on. If you’ve been both, a year apart, we should talk today.

What we do

Prompting stopped being the hard part a good while ago. Everyone does it now. The hard part is the step after: building the structure that lets autonomous models like Claude Fable and GPT-6 plan, build, test, deploy and ship a complete application, with no human writing code, and with code, architecture and tests of a quality no developer has the stamina to keep up by hand.

That is TokenTek’s whole job. We have migrated millions of lines of legacy code that way, fully autonomously. Independent external code reviews have given the code and architecture our structures produce their highest rating, and we show you the reports in the very first interview. Nobody here writes code by hand. When something is manual, we build the workflow that does it. The tools change with the client. The way we work doesn’t.

We have shipped apps with millions of downloads and a banking app with a million users. Our systems run at companies in finance, automotive and shipping. You would work forward deployed: in the client’s team, with their leadership, from first idea to running system.

What you would do here

• You design the controls the model can’t cheat its way through: the specification, evals, property and differential tests, and the gates between build, deployment and release. When the model writes both the code and the tests, it’s your job to make sure the tests mean something.
• You decide what each model is used for, what goes into its context and what it should fetch, and you’re the one reading the traces when a run goes wrong.
• You run legacy migrations of millions of lines, autonomously, and you prove that the result is correct.
• You build the data layer: retrieval, embeddings, vector search and knowledge graphs, exposed as tools and endpoints the client’s systems can call, with evals that show it works.
• You own the quality. What you deliver is built to pass an external review, not just a demo.
• You sit with the client’s team, understand the real problem and take it from vision to production.
• If you like teaching, there’s room for that too. Together with Hubbau we run a twelve-month AI Engineering Program for new graduates with a master’s degree, with our engineers as coaches.

What we need from you

• M.Sc. in computer science, systems architecture, informatics or similar.
• At least two or three years of experience building production software, ideally also with data integrations or ML models. There is no upper limit.
• That you can show something you’ve built with AI where the model did the work and you built the structure. A repo, a demo, a workflow. Ideally something with real users.
• Opinions. You know how models and harnesses differ, and you can argue for your choices.
• Judgment. Models will hand you working code all day long. You know the difference between working code and good architecture, and you insist on the latter.
• Swedish and English, spoken and written. Several of our clients require both.

What you get

• This job, all day, on the company’s account instead of your own.
• Colleagues who read the same release notes as you and are passionate about AI.
• Exciting projects where you can grow in a small, specialized team.
• Generous vacation, pension and insurance benefits, plus team trips worth closing your laptop for.

Practicalities

Full-time, based at our office in Göteborg. We mostly work from the office, with flexibility when you need it.