Nobody knows how to build with AI yet

Nobody knows how to build with AI yet
worksonmymachine.substack.com

This would take time to write if I’m doing it myself so I decided to vibe code it entirely. I had this idea that a compiled language is less likely to have errors (on account of the compiler giving the LLM quicker feedback than me) and so I chose Tauri with TS (I think).

The experience has been both wonderful and strange. The app was built by Claude Code with me intermittently prompting it between actual work sessions.

What’s funny is the bugs. If you ever played Minecraft during the Alpha days you know that Notch would be like “Just fixed lighting” in one release. And you’d get that release and it’d be weird like rain would now fall through glass.

Essentially the bugs are strange. At least in the MC case you could hypothesize (transparency bit perhaps was used for multiple purposes) but this app is strange. If the LLM configuration modal is fixed, suddenly the MCP/tool tree view will stop expanding. What the heck, why are these two related? I don’t know. I could never know because I have never seen the code.

The compile time case did catch some iterations (I let Claude compile and run the program). But to be honest, the promise of correctness never landed.

Some people have been systematic and documented the prompts they use but I just free flowed it. The results are outstanding. There’s no way I could have had this built for the $50 in Claude credits. But also there’s no way I could interpret the code.

Second of all, it's easy to fart out some program in a few days vibe coding. How will that fare as more and more features need to be added on? We all used to say "Dropbox that's just FTP wrapped in a nice UI anyone can make that". This protocollie project seems to be a documentation viewer / postman for MCP. Which is cool, but is it something that would have taken a competent dev months to build? Probably not. And eventually the actual value of such things is the extensibility and integrations with various things like corporate SAML etc.

Will the vibe code projects of today be extensible like that, enough to grab market share vs the several similar versions and open source versions anyone can make in a few days, as the author suggests? It can be hard to extend a codebase you don't understand because you didn't write...

I had stumbled upon Kidlin’s Law—“If you can write down the problem clearly, you’re halfway to solving it”.

This is a powerful guiding principle in today’s AI-driven world. As natural language becomes our primary interface with technology, clearly articulating challenges not only enhances our communication but also maximizes the potential of AI.

The async approach to coding has been most fascinating, too.

I will add, I've been using Repl.it *a lot*, and it takes everything to another level. Getting to focus on problem solving, and less futzing with hosting (granted it is easy in the early journey of a product) - is an absolute game changer. Sparking joy.

I personally use the analogy of mario kart mushroom or star; that's how I feel using these tools. It's funny though, because when it goes off the rails, it really goes off the rails lol. It's also sometimes necessary to intercept decisions it will take.. babysitting can take a toll (because of the speed of execution). Having to deal with 1 stack was something.. now we're dealing with potential infinite stacks.