Open role

    Head of AI

    At Caire, AI agents implement features from PRD to reviewed pull request in a monorepo with a merge queue and quality gates. As Head of AI you develop the agent harness, own the delivery pipeline from PRD to production release and optimise it for lead time, cost and quality.

    Apply with your GitHub

    Why the role exists now

    Caire has been built by one founder orchestrating AI agents. The agents handle implementation, tests and documentation, while the founder prioritises, reviews and approves releases. That has produced a complete planning platform for home care in a short time, but the delivery pipeline still depends on one person.

    We scale compute before headcount (compute-first). As Head of AI you are hands-on: you develop the harness, tooling and quality gates, ship to production and identify where human expertise is required.

    Why Caire

    What you get with us

    • Work that makes a difference

      What you build decides whether a client gets their visit on time and whether staff get through their day. The work shows up in the daily lives of people who need help at home.

    • Build a compute-first organisation from scratch

      At Caire, compute is hired first. There is no organisation of employees to fit into yet. You help shape it from scratch, in a company where AI agents do a large part of the work.

    How we know it is working

    This is what should be different once you have been here a while, and how we measure it.

    1. After 14 days

      You run the full pipeline from PRD to production, have established a baseline for the delivery metrics and have your first harness improvements in production.

      How we measure: Lead time from PRD to merged PR, model cost per PR and review rounds per PR, before and after your changes.

    2. After 30 days

      The delivery pipeline no longer depends on the founder. A feature goes from PRD to production without the founder implementing or reviewing the code.

      How we measure: The share of PRs that reach production without the founder, plus lead time for changes, cost per PR and change failure rate against the baseline.

    3. After 60 days

      The agents ship to production with fewer manual interventions, and the capability gaps that cannot be automated are mapped.

      How we measure: The share of PRs that reach production without manual fixes (first-pass yield), and a hiring plan for key roles, for example testing and AI quality or information security.

    4. After 90 days

      The delivery pipeline is complete. Anyone in the organisation can write a PRD from our template, and the harness takes it all the way to production.

      How we measure: The share of PRDs from people outside engineering that reach production, lead time from PRD to production release and change failure rate.

    What you own

    • Hands-on development in the platform and the harness, including your own production deployments.
    • The agent harness: orchestration, agent roles, model routing, tooling, context engineering, evals, agent-legible observability and quality gates in CI and the merge queue.
    • Delivery metrics for lead time, cost and quality per PR, including DORA metrics, visible to everyone every week.
    • Architecture and technical direction, together with the founder.
    • Code review as second reviewer, and vulnerability management.
    • Capability gaps that automation cannot close, and hiring key roles such as a testing and AI quality lead and an information security lead (data processing agreements, ISO 27001, row-level security and role-based access control).

    What is hard about the role

    • There is no existing engineering team. You start from the founder, the agents and an extensive codebase.
    • A defect in a home care schedule affects clients waiting for their visit. Delivery speed can never come before quality.
    • You decide when a problem should be solved with automation and when it needs a new role, and you own the trade-off.

    Tech stack

    Language and codebase
    TypeScript in a monorepo with Node 22, Yarn and Turborepo
    Web
    React 19, Vite, Tailwind CSS and Radix, with Bryntum Scheduler Pro in the planning calendar
    API and data
    Apollo Server 5, GraphQL 16, Express, Prisma 7 and PostgreSQL
    Mobile
    Expo and React Native
    Optimisation
    Timefold for route planning and staffing
    AI in the product
    Azure OpenAI in Sweden Central for customer data, and an MCP server for customers' own AI assistants
    Sign-in and operations
    Clerk, AWS in Stockholm, Sentry and CloudWatch
    Testing
    Vitest and Playwright
    Agents
    Claude Code, Codex and Cursor, with checks before every merge and a merge queue

    How we build the delivery pipeline

    1. 01Harness engineering: when an agent gets stuck, that is an environment problem to fix in the harness, not a task to take over by hand.
    2. 02The repository is the system of record. Specifications, architecture decisions and agent instructions (AGENTS.md and CLAUDE.md) are versioned and updated whenever an agent makes a mistake.
    3. 03Architecture rules are enforced mechanically with linters, structural tests and contracts, not with review comments.
    4. 04Verification is the biggest lever. Every change has a feedback loop of tests, evals and in-browser verification, and a separate agent reviews what is generated.
    5. 05Context is a finite resource. We plan first, decompose the work and leave handoff artefacts so that long-running sessions can resume.
    6. 06Throughput and stability are measured together. AI raises the pace, but without a stable platform it also raises the failure rate.
    7. 07Code health is maintained continuously with scheduled agent passes for bugs, refactoring, documentation drift and performance.

    We build on, among others, OpenAI's work on harness engineering, Anthropic's on context engineering, long-running agents and evals, Boris Cherny's Claude Code workflow and DORA's 2025 report on AI-assisted software development.

    Must have

    • Solid hands-on experience of production software development, including deploying and operating what you build.
    • Proven experience developing with AI coding agents in your daily work, visible in code you can share.
    • Experience building or extending an agent harness: orchestration, context engineering, tooling, evals, quality gates or CI/CD for agent-generated code.
    • Experience measuring and improving a delivery pipeline with metrics such as lead time, cost, change failure rate or review outcomes.
    • The ability to write precise specifications and acceptance criteria. Agents and engineers work from what you write.

    Nice to have

    • Experience from a leading AI-first company, for example a frontier lab, Lovable or Legora.
    • Experience with parts of our tech stack, such as TypeScript, React, GraphQL and PostgreSQL.
    • Experience with optimisation and scheduling (for example vehicle routing) or other decision-support systems where people review and approve the algorithm's proposals.
    • Information security and compliance in healthcare or another regulated field.
    • Leadership experience. It is a merit but not the core of the role, and leadership may also sit in another role.
    • Swedish.

    You do not need everything on this list. Apply if you meet the must-haves.

    Apply with GitHub

    We assess what you have built. Send a link to your GitHub profile. You do not need LinkedIn or a CV.

    Apply with your GitHub
    • A link to your GitHub profile.
    • Two or three repositories or PRs you would like us to look at. Briefly describe the problem, the technical choices, the lead time and how you verified quality (tests, review, measurement).
    • If you have built with agents: describe your harness architecture, how you verify the agents' work and the cost per delivery.

    Is most of your work private? Describe it, link a public demo or send a technical write-up.

    How hiring works

    1. 01

      Application

      You send what we ask for above. The founder reads every application, and everyone gets an answer.

    2. 02

      First conversation

      Thirty minutes with Björn Evers about you, the role and Caire. You get the salary range before the call.

    3. 03

      Technical walkthrough

      You walk us through code you have written: architecture, trade-offs and verification. Then we discuss how you would measure and reduce lead time and cost per PR in our delivery pipeline.

    4. 04

      References

      Two people who have worked closely with you.

    5. 05

      Offer

      You get a written offer with salary, terms and start date.

    Terms

    • Full time, permanent employment.
    • You get the salary range before the first interview.
    • Office in Nacka. We work hybrid.
    All open roles