Open role
AI
AI Engineer
Stockholm · Primarily onsite · Early stage
The Company
Togy’s mission is to double humanity’s output, and we start where output is made: the factory.
Factories were built to produce. Togy makes them think. Our product, MAIN, is an intelligence layer on top of the factory that already exists. It connects to the factory’s machines, systems and documents, and its AI coworkers answer questions, plan work and act alongside the people who run it, always with their approval.
We build MAIN at MagasinX in Stockholm, a former Ford factory, and serve our US customers from New York. We’re early and the problem is large, so everyone here owns a real part of it.
The Spec
Our AI coworkers plan and act on live production systems, where a wrong answer costs a shift’s output. You’ll make sure they get it right.
MAIN’s AI coworkers are the factory’s brain: they remember what the factory knows, answer from its data, documents and history, work together on a task, and act with the operator’s approval. You own that brain, from how a single coworker reasons to how many of them share memory and hand work to each other, and you keep it at the edge of what the field can do.
Most AI is built for chat and documents. A factory adds time series, alarms, decades of manuals and people who need an answer mid-shift, often on hardware inside the factory with no cloud to lean on. That’s what makes the work hard, and why it matters. You’ll spend time with the operators who tell us when an answer is wrong.
The Directives
- Own the AI coworker experience, from the question an operator asks to the tool calls, the approval and the answer.
- Own the harness around the AI coworkers: the tools they can call, the context they get, and how they recover when something fails.
- Design how AI coworkers work together: which one takes a task, how they hand work to each other, and how they share what they know.
- Build the factory’s memory at company, team and person scope, where people approve what it keeps, correct it and let it fade.
- Bring the right context to every answer: retrieval over manuals and SOPs, a knowledge graph with hybrid search, and summaries of raw telemetry a model can use.
- Make the AI coworkers as good on the open models a factory hosts itself as on frontier models in the cloud. Measure the gap, then close it.
- Build the evals and traces that tell us why an AI coworker got something wrong, and run them on every change.
- Follow the field closely, try new models and ways of building agents, and bring what works into MAIN.
The Stack
- Frontend
- React, TypeScript, Tailwind
- Backend
- Rust, Python
- AI
- frontier models in the cloud, open models on the factory’s own hardware
- Infrastructure
- Kubernetes, in the cloud and on the factory floor
- Cloud
- GCP, Cloudflare, Terraform
- Tooling
- coding agents in every part of the work
The Requirements
- You have built agentic systems that people rely on, not only demos.
- You follow the field closely and can tell a lasting idea from a passing one.
- You treat AI as software with higher entropy, and write typed, tested code that holds up in production.
- You write the eval before you trust an AI coworker, and you trust the numbers over a good feeling.
- You are comfortable making models reason over sensor data, time series and technical documents, not only chat.
- You use what you build: MAIN’s AI coworkers are part of your own working day, and the gaps you find become the next thing you fix.
- You work with coding agents every day, extend them with your own tools, and automate your own work first.
If this sounds like you, apply and tell us about something you built that you’re proud of.