- AI labs want real software work to train coding agents: code with its history, issues, reviews and the discussion around them.
- A closed crypto startup received about $25,000 for its codebase, with a workspace data sale of about $15,000 pending, per MIT Technology Review.
- Open-source components keep their own licenses, and secrets and customer data in repositories must be removed.
- Live companies can license historical data without giving up current code.
What engineering data is worth to AI labs
Coding agents learn from how software is really built: commits, pull requests, reviews, bug reports and the conversations that resolve them. That is why buyers want history, not just a final snapshot. See source code datasets, software issue history and agent trajectory data.
MIT Technology Review reported a closed crypto startup receiving about $25,000 for its codebase, with workspace data likely worth about $15,000 more (MIT Technology Review).
Traps to check
- Open-source components. Code you did not write keeps its own license. Exclude vendored dependencies or list them with their licenses.
- Secrets. Repositories often hold API keys, passwords and tokens in history. Scan and remove them before anything leaves your systems, and rotate them anyway.
- Customer data. Fixtures, logs and support attachments may contain customer records. Remove them.
- Contractor code. Check that contractor agreements assigned rights to you.
- Third-party platform terms. Exports from collaboration tools may be restricted by their terms. See Slack API terms.
Live companies
You can license historical data (for example, a deprecated product or code older than two years) without exposing current work. Restrict the license to training, prohibit reproducing your code verbatim, and require deletion on termination.
Closing companies
If you are shutting down, code and workspace data may be among your most valuable remaining assets. Read selling data from a closed company and see SimpleClosure.
Read next
Questions
Can I sell my startup's source code to an AI company?
Yes, if you own it. Exclude open-source dependencies or list their licenses, remove secrets and customer data, and confirm contractors assigned their rights.
Do AI labs want our Jira or GitHub issues?
Issue history with linked code changes is valuable for training coding agents, because it shows problems and how they were solved.