A.I code editors are great. First I started using Tab 9 and it was great. Then I started using WindSurf, still use it, I just use it for autocomplete, because it gets spent easily if you use it for bigger tasks. Then came Kiro, its plans were more liberal, and then Antigravity with more liberal plans than Kiro.

Every time I tried these new tools, I went: “oh my God!”, it was that Good. Sure I can see humanity improving fast. A.I is going to be cheap, super cheap, because companies like QANT, Extropic, and so on are coming up with novel ideas, energy cost of training and running A.I is going to come down dramatically. Soo A.I code editors will become like email, all of A.I will become like free email. We can use it for free, while Google or any other evil corp, makes a ton of money with our data.

Anyway, let me tell you what features I want in my near future A.I editors:

Air draw

Today I need to sketch out in tools like excalidraw, and upload an image to the A.I, telling it what I want. But what if I can press a button, my camera gets switched on, and I can draw on air, and the diagram is shown on my screen, once the diagram is done, I can press upload button, so that what I drew is sent to A.I model. This feature can be implemented very quickly by any editor these days.

Voice input

I need to be able to talk with the A.I, than type, common, it’s 2025, we must have her. Atleast as coding assistant.

OS integration

A.I need to be integrated with OS, yes there is MCP and other stuff, but I think there must be a basic optimized OS integrated A.I. This will not only make me code fast, but make stuff like launching and testing apps a bit faster. We need a A.I OS. I wish Ubuntu GNU/Linux leads here, than Mac, or Windows which both are crap systems to code with.

Binary generation

Let’s say I write a code, why can’t I have binaries generated for GNU/Linux, Mac, Windows, Android, iOS and other OSes generated? A.I can do it. It should also package the code into installers with one command. Possibly launch it into app store.

Today its only possible to write software for Apple devices only on a Mac, you need a Mac to compile it. But A.I can bypass that in near future. It can simply reverse engineer entire Apple ecosystem, and turn what ever code you write for say Debian GNU/Linux, into a Mac compatible binary too.

No cloud, locally working

Well, this is not A.I coding specific. I don’t want to be tracked, I want cheap enough hardware that can be plugged into my laptop, or over the WiFi, running something like Gemini 3. I think its possible by pruning models, or by compressing the model. Samsung has invented a technique to compress 30B parameters to 3B. There are recursive models that beat Chat GPT, with just millions of parameters. So I think its possible.

Why do I need human understandable form of code?

Why do I need human understandable code? It’s lengthy, and yuck. I just need a software that needs to run on my computer. So if near future development environment just generates binary then it will be cool.

Testing

I think A.I should also test software well. Antigravity integrates well with Chrome. But Chrome is a spy, that takes data of our activities and tells to Google. I want to use browsers that care about humanity, like Firefox. Plus A.I agents should emulate Android, iOS, or, whatever OS in my computer, and must test the binaries. I think that’s possible.