That's only half the author's point. The other half is that the "gate is where it always was" (= the part that's not just grinding out code to spec), and the fact that you can't vibecode a Photoshop doesn't mean AI is useless.

Indeed. But:

> the fact that you can't vibecode a Photoshop doesn't mean AI is useless.

I don't even think this "can't" is even a fact at this point.

24 years ago, my first major academic project, I wrote myself an image editor. In Visual Basic, because the teacher required it, and without version control because I hadn't heard of that when I was 18. It wasn't Photoshop, more like PaintShop Pro without the plugin support and instead with a bunch of baked-in effects, but the experience (and a later attempt to make it into a useful product) showed me how easy it was to separate concerns in that kind of app*, so I think it would be possible if anyone actually cared to try it.

* for each document, you have a fairly simple data structure in the form of an array of layers/groups, each layer has masks etc., but mostly you're building out a huge number of other functions on that data; the hard parts are performance, which AI can also optimise; and that some of these (e.g. smart selection, content aware fill) need some kind of AI to be any good, which again AI can also now create the training system for.

Even re-implementing a JPEG codec from scratch in a stupid language for the task wasn't too painful for me while I was still a university student, and I only did that because my REALBasic-on-a-PPC-mac setup at the time didn't allow me to use libjpeg like a sensible person.

Of course, Adobe also has Creative Cloud and a search engine over stock photography, both of which are essentially entire projects in their own right, and I'd assume also integration with all their other apps.

Proper vector fonts were also out of my scope; that and full backwards compatibility with PSD format, undocumented warts and all, would be my only real question for a vibe-coding attempt.

I think its also possible, but non-technical people likely wont hill climb high enough to make this unless they are extremely motivated. Using abstraction properly to effect change, understanding breaking down the problem to extreme details and building them back up one by one. I think most of the people using AI for one shot type of things don't even know what "content aware fill" is even if they used it.

Most vibecoders say their stack is "cursor". Reread your comment and then try to guess if a vibecoder could even understand what you talked about let alone execute it in a pefect and reliable manner. Vibecoding is about dopamine hits, people give up if it doesn't work on first 2-3 attempts.

> and the fact that you can't vibecode a Photoshop doesn't mean AI is useless.

A lot of ai hype IS premised on ai being able to vibecode photoshop.

I never heard of it, let alone "a lot". Care to give examples?

Photoshop (and many traditional SaaS products) solve hundreds of different use cases. Most users probably only care about a handful of them. You don't need to do every use case to kill SaaS if you have a tool that can allow users to solve their 2-3 use cases on their own with custom tooling.

And the point of the comment you are answering is that the market you are talking about has taken on a different form.

The difference is that normally, people would perhaps pay a company to buy a PS license, or pay a professional to edit their images, however now this market segment will just use VLMs to edit the image to what they need.

It goes beyond that. It's becoming cheaper and easier to e.g. vibecode tools to carry out the handful of uses lower end users may have for PS. Some users will do that directly. Some users will turn that into services or apps. PS will be weakened by thousands of small cuts, not one large vibecoded abomination.

Some of those tools will use VLMs to provide more advanced features instead of implementing it themselves, making competing for a broader subset of users feasible.

What is "VLM"?

Vision Language Models

Very large models that aren’t large language models.

1. Adobe sells Photoshop to design professionals.

2. Design professionals sell their work to businesses.

But if businesses start using AI image editing instead of contracting professionals, then Adobe won't have a market of Photoshop buyers.

Before you say that AI isn't good enough for the kind of business which pays for professional design, in one year it probably will be.

Its not about AI being good enough. Businesses outsource these things because they don't want to deal with it. That's the same now. The question is if designers are using AI or not. And guess who has the best AI stack for designers? Photoshop with Firefly.

This is why I insuniated that people here don't really understand photoshop as a business and customer, only see it as an app.

And yeah, it will be better in one year but so will Adobe's inbuilt AI tools.

We have not seen designers abandon Adobe for AI and i am willing to bet that wont be true, not even in this decade.

You're making really good points.