I recently saw some chaps rebuilding the entire adobe software stack with AI over at https://getartcraft.com/apps. At a first glance it seemed pretty interesting, but then I started wondering about the economics of it all. I took out my imaginative napkin and started crunching maths (if you can call them that, I basically guessed every fucking variable).

So lets start with lines of code likely contributed directly by AI in the range of September 27th to October 9th.

I pulled these numbers out of my ass out of the contributor page for each of these projects. I looked at the top 5 contributors there and picked the most likely people that would just tell AI to do all the work. Oh, and I also included the guy named Claude. Must've been tough out in San Fierro, I respect his turn over to software engineering. In any case, we are looking at 3672848 lines of code of straight up AI generated... slop? goodness? You decide, cause I'm biased. Next since the project is written in rust, we can assume an average of 50 tokens per line. This time I did pull the number out of my ass, but I also took into consideration the fact that rust is a complex language and might require a lot of tokens to compensate. With that we can get to the number of output tokens, which would be 183642400. We can probably very generously assume 250000000 input tokens, which would include all the research, out loud thinking, other yapping and just rereading the same code over and over again (because clankers have no memory to speak of). Now we have to figure out which model was used. I can't be fucked to do investigative research, so I'll just assume they used claude opus 5.5 due to it's relatively decent performance and low cost (as opposed to fable 5.1). To be clear I did see codex over in the contributors, but claude was consistently on top. After all that guesswork we can finally get to the part where we do actual math.

To recap: 250000000 input tokens, 183642400 output tokens, claude opus 5.5 (4 usd input/m, 20 usd output/m), caching I'm not even going to bother so assume it's the worst case scenario. We are looking at 1000 USD for input, 3673 USD for output, landing us at 4673 USD for the entire suite of tools. You'd think it's not that bad, but these projects are years away from being production ready and they are officially in their alpha testing stage. Let me paint the picture even worse for you, because without the generosity of nearly trillion dollar companies you'd be looking at AT LEAST twice the price for the same api calls, meaning you're looking at 8000 USD already. Their generosity will run out, and it will run out soon, meaning no 20 dollar subscriptions with god knows how much usage packed inside of it and no "cheap" api calls. None of that. If you ask me, this looks like yet another attempt at a PR win by these AI companies, just like with bun. Too bad bun is fucking dogshit now, though, maybe they'll succeed with this? Because we are sure as hell not done with the ai companies trying to manipulate the publics opinion on them. ("The Fake Accounts Fueling OpenAI’s Political Machine" by Midas Project)

Regardless of all that, people still root for the project because of the overwhelming amount of reasons to hate adobe (Click to see even a friend of mine being happy to see adobe getting challenged). Which is fine, fuck adobe and especially Shantanu Narayen.

Abuse AI while you still can, and preferably do it for free (or for dirt cheap), because it ain't worth dick compared to the thing inside of your skull. ;)