Changed my mind on vibe coding

Build in public day 28/30: I changed my opinion on vibe coding today. It can help you move extremely fast under certain conditions. Will show a couple things off later this week.

The requirements today seem to be having relatively small codebase with well-documented frameworks/libraries (e.g. Laravel and React) and solid guardrails (E2E tests, plenty of examples, etc.). This will change as models improve.

That takes expertise up front to plan/architect things right and the ability to dig in and fix things when they go off the rails, under the right conditions you can really move 10X as fast.

Vibe coding will continue to get a bad rap as novice users get in over their heads, but over time a few things will happen:

1. New boilerplate kits will emerge with the right setup/examples/foundations/prompts to better keep AI on the rails.

2. Services built specifically to help fix/undo vibe coding issues and make it more fault tolerant (like an intelligent version control e.g. w/smart branching).

3. The AI will continue to improve and raise both the ceiling and the floor.

As AI improves today's coding will feel like writing assembly in a few years. Why? Because technology moves ever upwards towards higher levels of abstraction. It always has and always will.

It'll still take a while for it to be good enough for the enterprise broadly since large companies have so many undocumented frameworks/libraries, multiple versions, complex codebases, etc. But that'll also get solved in time, especially as 'enterprise software' as we know it goes away.