Uploaded August 2026 | Updated September 2026, 3 weeks ago

---

In this episode, JK Gunnink walks Martin Omander through his actual, hands-on workflow for tackling poorly documented legacy code using AI.

No theoretical hand-waving here. JK shares the practical engineering strategies he uses to safely modify legacy systems without breaking things, including things like:
* Preparing the sandbox: Why documenting your local build, linter, and testing setup in a README is the ultimate prerequisite for letting AI agents self-correct.
* The model split (BDD): Using pro-tier models to write strict acceptance tests first, then handing execution off to faster, cheaper flash models to write the implementation until the tests turn green.
* Preventing context drift: Keeping the agent focused by breaking massive tasks down and maintaining a running markdown checklist.

Chapters:
0:00 - Intro
1:35 - The plan step
4:00 - The execute step
5:54 - The verify step
6:31 - Takeaways

Watch more Serverless Expeditions → https://goo.gle/ServerlessExpeditions
🔔 Subscribe to Google Cloud Tech → https://goo.gle/GoogleCloudTech

#ServerlessExpeditions #GoogleCloud

Speakers: Martin Omander, JK Gunnink
Products Mentioned: Gemini