Plausible premise, and it will be interesting to see what happens.
Two things though:
- I don’t trust any big tech companies to have their head on straight when it comes to AI in software development
- Seems like “2x the time” is a cop-out argument. Hands on keyboards is surely not a limiting factor when you’re a 10$bn+ company. I imagine it’s more a problem of coordination and source-of-truth. Those problems don’t go away with AI. If anything, they get worse. Neither the iOS nor Android app is the “authoritative” version, so are they going to have AI copying back and forth between the two, over and over? Seems like the same kind of “localized model collapse” problem lots of companies are sprinting towards.
For me this was an interesting read. I liked the approach to make core business logic accessible through a CLI instead of just UI. This makes testing a lot easier, no matter if you are using an LLM to execute/create the tests or a conventional approach.
It’s pretty good practice to have your app modularized to the point where you can execute the core business logic separated from your UI layer, for which you then don’t even need a CLI.
In the same vein, it’s good practice to have your UI be modularized to the point where you can test its logic separated from the core business logic
I agree, every program should be built this way. It also speeds up development early on, and makes it easier to have multiple front ends.
TL;DR: it‘s because LLM agents do the work now.
Given that its CEO has some serious issues with democracy, I can’t say that I care one iota which language abomination is powering his empire.
