Prediction from the AI-max timeline: Vibeshop

What if Team AI bubble is wrong, and thesuccess of LLMs in filing security bug reports is an early sign that we’re on the AI maximalist timeline after all? What if all of these are true?

• The capabilities of “AI agents” keep going up. Software projects will be able to do something like Tim Sehn’sA Week In Gas Town, but for more categories of software, including applications with featureful GUIs.

• Big AI wins copyright and related cases. On the AI-max timeline, politicians agree with Anna of Anna’s Archive thatCopyright reform is necessary for national security, and while they’re at it, they nerf a bunch of content-related contracts including licenses, EULAs, and ToSs too. Maybe even patents.IP Maximalism continues to go out of fashion. So far it looks like Bright Data has some powerful friends, and is winning in court.

• Inference and other runtime costs stay reasonable or come down. Theusage-based pricing problem turns out to be a nothingburger, the pricing Time Bomb detonation turns out to be just a blip, competition and technological advances drive prices down, all is well for vibe coding budgets. Maybe Ben Thompson’sInference Shift happens, maybe something else?

All of those would need to happen, but what if they do? On that timeline, things would get interesting in a lot of ways. One of the biggest would be, let’s call it “Vibeshop,” the AI-built clone of Adobe Photoshop—or more like clones plural, since the project could make a variety of builds, each capturing the points in Photoshop’s history when some group of artists really liked it.

In Adobe Has Rid Itself of Its Allies, Nick Heer explains the Creative Cloud product direction, which helps explain why Photoshop is a good target for a vibe clone.

I think Adobe has actually shipped worse products as a result of this strategy — and, for once, I will avoid making it all about bugs, of which there are many. Adobe’s applications are more capable than they ever have been, but they are also often worse for professionals in actual use as a direct result of the company’s software-as-a-service model.

Users have to turn on Quiet Mode to avoid upsells. And Adobe’s ‘Modern’ User Interface Is Just Webpages.

Another reason Photoshop would be a good early target for an automated clone would be all the existing books and training videos that could be turned into test cases. (Remember, this is on the timeline where copyright and licenses get weakened because AI lobbying and/or competition with the PRC.) The Vibeshop “agents” could run a pre-ACC Photoshop in a VM, work through the tutorial, and keep tweaking their own code until results match. (Yes, this is also the timeline where LLM APIs get cheap.)

A company built around Vibeshop could offer the bleeding edge releases for free, plus supported releases of particular setups for artists who are used to having things one way and don’t want to change. And yes, the company that sells subscriptions might not be the same company that runs Photoshop in a VM in a license-violating way, but in a world whereGoogle can make ad revenue from obvious infringing sites, we know that it’s possible to stand up enough insulating layers to make it work.

Of course, Adobe could always do Vibeshop to themselves, as Sun Microsystems could have done if they acted on Larry McVoy’sSourceware Operating System Proposal instead of letting Linux eat their lunch. The existence of Vibeshop could turn into a long-term political asset for the company and the industry, too. One reason that firearms manufacturers have a loyal political constituency and IT companies don’t is that in the firearms business, management is smart enough not to periodically alienate the best customers. A customer can walk into a modern gun shop and buy a product that could have come from John M. Browning’s workbench more than a hundred years ago. What would happen if the same were true of software? (More:Learning from Second Amendment defenders)