How we work with AI

We develop agentically with Claude Code. That lets us deliver much faster - but every line of code we hand over has been reviewed by a human.

Requirements first. Before the first prompt, we agree with you on what the app must do and how we'll know it's done.
Who does what. The agent writes most lines of code. We make the decisions: requirements, architecture, review of every change and sign-off on every release.
Rules and guardrails for the agent. Every project has AI rules in the repository (CLAUDE.md): architecture, conventions and what the agent must not do. Automated hooks block it from writing to production and from reading passwords and keys.
Risk-based review. No code reaches production without human review. Changes to data, permissions and calculations are reviewed line by line and tested by hand.
Tests and continuous delivery. Every project has automated tests and CI. You see changes on a preview version before they go to production.
You own the code from day one. The repository lives in your organisation - no vendor lock-in.
Your data stays under control. We don't send sensitive data to AI tools and use business plans that don't train on your data.
We hand over the AI context too. Along with the code and documentation you get the rules and skills for the AI agent, so anyone can take the project over - with or without AI.
GIS is done by a human. Coordinate systems, data quality and domain logic are Petr's job - no model solves that for us.