Agentic engineering for product teams
Agentic engineering means AI agents do most of the coding inside your real codebase, while your team steers, reviews and approves. We set up the layer that keeps it reviewable and under control — rules, skills, specialist and review agents, quality gates and a spec-driven workflow — and we keep it running.
The founder works this way as a contributing engineer on Sweden Startup Next and Ignite Nordic. What we install is set up for your codebase; none of it comes from those teams' tooling.
For teams that want AI to carry real work.
Product teams of 2–10 developers
You want AI to carry most of the coding without loosening review, and nobody on the team has set it up before.
Startups without an engineering lead
Early-stage companies that need review discipline and quality gates before their first engineering hire.
Teams handling identity or personal data
BankID, payments or personal data in the product, and AI features that must ship with an access model and documentation.
The agentic layer, piece by piece.
Everything below lives in your repositories and CI, yours to keep and change. Nothing depends on a service we host.
Rules
A written rules file the agents must follow: your conventions, your security constraints, your definition of done.
Skills
Reusable, versioned instructions for recurring work — debugging, testing, migrations, reviews — so agents don't improvise.
Specialist agents
Agents with one job each, chosen from what the audit finds — for example a database reviewer for a Supabase codebase, or an accessibility checker for a public site.
Review agents and gates
Every pull request is checked for the kinds of defects your audit found most likely, such as data-access mistakes or regressions, before a human looks. CI gates stop whatever fails.
Spec-driven workflow
Tickets become specs with acceptance criteria before code is written. Every fixed production bug becomes a test or a rule, so the same bug is caught next time.
Developers steer, review and approve.
- Developers spend their time steering, reviewing and approving rather than hand-writing most of the code.
- Every ticket needs a definition of done before an agent picks it up. We help you get the backlog there.
- Nothing merges without a human approving it.
Where we draw the line.
- We don't install anything you can't keep: it all lives in your repositories.
- We don't bring tooling or code from the teams the founder works in.
- We don't build AI systems the EU AI Act lists as high-risk, such as creditworthiness scoring of individuals, recruitment and selection, or admission and grading in education. Their obligations are outside our scope.
- We don't give legal advice. Our compliance work is technical preparation and documentation, for your lawyer to review.
Three ways to start.
Every engagement starts with a free intro call. The audit is the first paid step, and its fee is credited against a build.
Intro call
Where to start.
Video call. We look at your repos, CI and backlog and give a candid read of where AI can carry work. Written summary the next working day.
- Repo & CI first read
- Backlog first read
- Written summary
Readiness audit
Planned on facts.
1–2 weeks. Codebase and CI review, definition-of-done scoring, security, RLS and GDPR gaps, build-vs-buy. The report is yours to keep, and the fee is credited against a build.
- Cap agreed before we start
- Written report
- Prioritised gap list
- Credited against a build
Build + Run
Ship it, then keep it working.
Agentic layer installed and one feature shipped through it, or one AI feature in production. Then operations: gates, agents, monitoring and incident response, with a 4-hour weekday response.
- Fixed module price
- Agentic layer or AI feature
- Monthly retainer after
- 4h weekday response SLA
- Monthly report
All prices excl. VAT.
Common questions.
- What is agentic engineering?
- A way of building software where AI agents write most of the code inside your real repositories, under written rules, review agents and quality gates, while developers steer, review and approve.
- Do we need to change our stack?
- No. The layer is installed in your existing repositories and CI. We have run it on TypeScript/React and PostgreSQL/Supabase with GitHub Actions; other stacks we check in the audit.
- Which AI tools do you use?
- Mainly Claude Code. Rules, skills and specs are plain text files in your repository, so they stay usable if you change tools.
- How long does it take?
- The readiness audit takes 1–2 weeks. For installing the layer and shipping the first feature through it we plan for 4–8 weeks, depending on the codebase and how well prepared the backlog is.
- What does it cost?
- The intro call is free. The audit is 900 kr/h with a cap of 15 000–25 000 kr agreed before we start, and the fee is credited against a build. Build modules start at 60 000 kr and the run retainer at 18 000 kr per month. All prices excl. VAT.
- What happens to our code and data?
- Everything we install lives in your repositories. We sign your data processing agreement before touching any environment with personal data, work read-only by default, and keep secrets in your vault, never on laptops.
Tell us what you want AI to take over.
Most conversations start with a free 30-minute intro call. No obligation, written summary the next working day.