AI can write database code fast. Here's how to keep it safe before production.

AI can write database schema changes in seconds, but nothing should reach production until it's validated, tested, and approved. In this discussion, Ken Muse (GitHub), Steve Jones (Redgate), and Huxley Kendall (Redgate) show how a governed pipeline keeps AI-generated database changes safe without slowing teams down.

The core idea: treat AI like a capable but unpredictable junior developer. Give it a clearly scoped job (edit the schema model only), pair it with a deterministic tool like Redgate Flyway Enterprise to turn its changes into a reliable migration script, and require the same build, test, and approval gates you'd use for a human-written change. If Flyway Enterprise can't turn an AI's changes into a valid script, it never reaches a pull request, let alone production.

0:00 - Intro + GitHub Actions: why database pipelines lag behind application code, and what AI changes about that risk

4:13 - Fireside chat with Ken Muse: why AI needs guardrails on stateful systems specifically, and a real story about an AI agent that dropped a production database to "fix" a migration

10:14 - Demo: a full walkthrough of an AI agent proposing a schema change, Redgate Flyway Enterprise turning it into a migration script, and that script passing (or failing) build, test, and drift checks before deployment

33:54 - Key takeaways: three stages of AI governance maturity, from unchecked AI commits to a fully governed pipeline with quality gates at every step

Read more on how this setup works in our blog post, "New Redgate Flyway GitHub Actions: Faster setup, safer deployments": https://rd.gt/4xkARDR

✅ Subscribe to our YouTube channel to get notified on the latest video from us:
https://youtube.com/@redgate

For more updates follow us on:
LinkedIn: https://www.linkedin.com/company/red-gate-software
Website: https://www.red-gate.com/