How I use AI
Last updated: 11 November 2026
I build with Claude Code, Anthropic’s coding agent. It writes much of the code. I read every change, run the tests, and answer for the result as if I had typed it myself. If you do not want AI on your project, I quote without it.
Your code is not used to train models. That is in Anthropic’s commercial terms, not a setting I promise to keep on.
Tools
- Claude Code. Client work runs only on Anthropic’s Claude Team plan, under Anthropic’s Commercial Terms, which include its data processing addendum.
- Not used on client work: free or personal AI accounts, or any tool not named on this page. A new tool goes on this page before it touches client work.
What AI does, task by task
| Task | The agent | Me |
|---|---|---|
| Reading your codebase | Searches and summarises the files I point it at | Decide what it may read |
| Design | Proposes options when I ask | Choose the design and write it down |
| Code | Writes most of it | Read every diff before it is committed |
| Tests | Writes tests from the agreed behaviour | Check they test the right thing, run the full suite |
| Review | A second pass over the diff | Final review, approve or reject |
| Handover document | Drafts it | Edit and sign it off |
| Estimate and price | Nothing | All of it |
| Emails and calls with you | Drafts some messages when I ask | Edit, send and stand behind every word; calls are mine |
One change, start to finish
- I write the task: what should change and how we will know it works.
- The agent proposes the change and its tests.
- The checks run: the build, the linter, the full test suite, and my own verification gates, which check the change against the task.
- I read the diff. If it is wrong, it goes back to the agent or I fix it by hand.
Your code and data
- What is sent: the files and command output the agent reads during a session go to Anthropic’s API.
- Where it runs: on Anthropic’s cloud API. Nothing runs on your systems unless we agree it.
- Training: Anthropic’s commercial terms: “By default, we will not use your inputs or outputs from our commercial products” to train models (Anthropic’s privacy centre).
- Retention: 30 days, Anthropic’s standard period for Team and API accounts.
- Location: Anthropic states that “data is stored in the US.”
- Never in a session: passwords, API keys and other secrets, and real personal data. I work with test data. I do not send feedback or bug reports to Anthropic from a client session.
- When the work needs personal data: we sign a data processing agreement first, and it names Anthropic as a sub-processor.
Ownership
- Once the invoice is paid, you get all the rights I hold in what I deliver.
- Copyright in AI-written code is not settled law. My review and edits are part of every change, but I do not promise that every line is copyrightable.
Your say
- The contract names the tools and what they may read.
- If your contract or your company’s AI policy is stricter than this page, yours applies.
- You can limit AI to parts of the project or rule it out. Without AI I quote more hours and a later date; the price is still fixed.
- If your company has an AI or supplier questionnaire, I fill it in.
Questions buyers ask
- Who is responsible for defects? I am, the same as for code I type; the warranty in the contract applies.
- Did AI write this page? It helped draft it. I edited it, checked every claim against its source, and stand behind every line.
Changes to this page
- 11 November 2026: first version.