The problem with (propietary AI) is that they (anthropic/google/openai/etc) gain more from the usage of AI than you. Other tools like postgres, gcc, git, HTTP, emacs, etc. don't "gain" anything if you use them (well, they gain popularity and perhaps more contributions, but that's it). The more you use Claude, the richer anthropic gets and the easier for them to position themselves in a place of power, power to dominate the programming of the world. That's sad. So even we all like so much propiertary AIs, we should think twice what we are giving in exchange (and no, it's not just the $200/month what we are giving)

I'm all for open models, open source agents, etc. I don't want to give more power to the big corps, though. Imagine what software engineering could become in 5 years if all thse big corporations gain even more power over us. It's a terrifying scenario (e.g., pay more so we don't show you ads in between claude code prompts; pay more so that the produced code doesn't incrust ads in your app...). Do you really want the same shitty experience we have now in the global internet, but deeply ingrained in your software engineering workflows?

There were lots of proprietary database companies in the 70s-90s when they first came out (Oracle, IBM, Sybase, SQL Server), but these days open source databases are the default. Everyone's making so many wild predictions based off the current state of LLMs. We don't know how the market will unfold.

Even in regards to programming competency and the culture of vibecoding, it's similar to the early stages of electric cars where they fill some roles better than ICE engines, but it's still a decade away from doing the same job. And the whole time there was people trashing electric cars as a novelty/impractical/expensive/dangerous/etc because the infrastructure wasn't there yet and the technology was immature.

The only real moat pattern we see is datacenters being hotly in demand but even those will scale up and commoditize, RAM manufacturing will catch up etc.