Tools and services
AI Tools
This page defines which AI tools are approved for use at Abtion, what data can go into them, and what we expect from AI-assisted work.
Approved tools
- 🟢 Claude - for chat-based assistance, reasoning, and writing
- 🟢 Claude Code - for agentic coding work (multi-step tasks, codebase exploration, scaffolding)
- 🟢 GitHub Copilot - for PR reviews and in-editor suggestions
Before introducing a new AI tool to a project, check with the Principal Engineer or CTO. Tools that send data to third-party servers require extra scrutiny for GDPR compliance.
What data can go into AI tools
Never send to external AI tools
- Personal data about users or clients (names, emails, CPR numbers, financial data, health data)
- Credentials, API keys, or secrets of any kind
- Client code or data that is subject to a confidentiality agreement, unless the tool has a confirmed data processing agreement (DPA) in place with Abtion
Generally acceptable
- General programming questions and patterns
- Anonymised or synthetic data
- Internal code where no client confidentiality clause applies
When in doubt, anonymise the data before pasting it, or ask your manager. If a client has specific requirements around AI use, those take precedence over this policy - document them in the project README.
Expectations for AI-assisted work
AI-generated code is held to the same standards as hand-written code. That means:
- You own the output. Committing code you cannot explain is not acceptable, regardless of how it was produced.
- Review AI output critically. AI tools hallucinate APIs, introduce subtle bugs, and sometimes over-engineer solutions. Treat AI suggestions as a starting point, not a finished answer.
- Keep it lean. AI tends toward verbosity. Remove anything that is not needed.
- Tests still apply. AI-generated code must meet the same test coverage requirements as any other code.
See the progression framework for how AI usage expectations scale with seniority.