3 Comments
User's avatar
JP's avatar

The second loop is the killer. Review fatigue hits hard when every PR is AI-generated and people just stop reading diffs properly.

Took a different angle on the same problem. Put together a pre-ship checklist so vibe coders can catch the security and infra gaps before code even reaches a reviewer: https://reading.sh/your-vibe-coded-app-works-now-ship-it-properly-80b98f77ec25?sk=cabfc3bc74dee0495f0772382bd9a70e

Mathew Jose's avatar

Vibe coding only started recently. Give it some time to mature and learn with it.

A junior engineer takes time to become senior engineer, right? And Tesla took close to a decade to gain market share.

MetalMonkey's avatar

Many, many thoughts on all this! How much quality control went on the genies training data? GIGO springs to mind. A personalised genie trained on just my/our/ABC Corp code might work but even then, what's the quality percentage there? TDD, TDD, TDD. No pass, no commit, do not pass go, do not collect £200. Different genies, different jobs, architecture, design, test etc so context isn't corrupted maybe. Certainly send the genie to silicon heaven if it gets into a denial loop!

As always, love reading your thoughts on these things!