Report · PDF · 26 pages
Your vendor uses AI to write your code. The 7 questions to ask them
The questions to ask them, the answer that reassures and the warning sign in each one.
2,500 → 30,000
changes per day in Anthropic's code, from May to September 2026
Anthropic, Oct 15, 2026
25%
of contributions reaching open projects are low quality
Linux Foundation Research, Oct 20, 2026
35%
of projects have a formal policy for code made with AI
Linux Foundation Research, Oct 20, 2026
What's in the material
Seven questions that fit into a progress meeting.
The seven questions below fit into a progress meeting. None requires technical knowledge to ask, and all have an answer that reassures and one that should worry you.
- 01Do you use AI to write our code? In which parts?
- 02Who signs off on each change?
- 03What has to pass before a change reaches production?
- 04How do you know the test tests?
- 05Where is it written why each decision was made?
- 06Does our code or our data go to any AI tool?
- 07If you leave tomorrow, can our team maintain the system?
2026 report · 26 pages · 9 public surveys
Get the full report
Five chapters with the numbers from nine public surveys, the seven questions for the vendor, a checklist, sources and a glossary.
How Code and Soul answers these questions
We use assisted generation every day, including to build what we sell. Each commit has a person named as author and the model named as co-author. A missing check is built before the fix, runs in the build the deploy calls, and is proven by breaking it on purpose. The reason for decisions is written alongside the change. And nothing is called done without running in the environment where it will live.
Sources
- Anthropic, Solving code review at Anthropic · read on Oct 15, 2026
- Linux Foundation Research, Navigating Open Source Software Development in the Generative AI Era · read on Oct 20, 2026
Photos: DC Studio, pvproductions, wirestock and Freepik / Freepik