The Story Everyone Is Telling

It goes like this.

Adobe charged people hidden fees to cancel their subscriptions, bad enough that the US government sued. Then it made everyone re-accept terms so vague that many read them as letting Adobe train AI on their work. Sony is ending physical discs for new games in 2028 and wouldn’t budge even after a petition passed 350,000 signatures. People are furious.

At the same time, AI got good enough. Not AGI, but good enough that a person with an idea and a weekend can ship an app. So surely the angry users just… build their own Photoshop? Their own Premiere? The revolution writes itself.

I wanted to know if that’s true. So instead of vibing an answer, I went to the evidence: earnings filings, randomized trials, labour data, court rulings. Here’s what came back.

Short version: the revolution is real, but it’s not the one you think.

Then You Look at the Numbers

Adobe’s most recent quarter, the one that ended in August 2026, brought in $6.76 billion. A record. Up 13%. Subscription revenue from individuals and business users grew 16%. That’s from their own SEC filing, not a press release.

That’s after the outcry. It’s also after Canva did something wild: in October 2025 it made Affinity, a genuinely professional Photoshop/Illustrator/InDesign competitor, free (only the AI tools need a Canva subscription). It hit 3 million downloads in 33 days.

A free, polished, pro-grade alternative. Massive anger at the incumbent. And the incumbent’s revenue still grew double digits.

Sony said the disc backlash is having “no impact on our business”. In fairness, that quarter ended before the announcement, so the real test is Sony’s next quarterly results. Watch that one.

And the vibe-coded replacements? There’s one high-profile attempt: ArtCraft, an open-source suite cloning Photoshop, Premiere and five other Adobe apps, released this month. Tellingly, its creator isn’t a weekend vibe-coder. He’s an engineer with 15 years of experience who used AI as a power tool, and he says so himself. It’s still in alpha. The Photoshop clone has around 45,000 GitHub stars, but stars measure curiosity, not studios switching. Nobody has shown it being used for real work yet.

The outcry is real. The exodus isn’t showing up.

Starting Is Cheap. Finishing Isn’t.

Here’s the part I find most interesting, because it’s about how AI coding actually behaves.

AI is genuinely great at the first 80%. Small project, clear spec, fresh codebase, and you’ll fly. Every study agrees on that part.

The trouble is the last 20%, and three findings tell that story better than any opinion:

1. Rebuilding real software from scratch still fails. A 2026 benchmark called ProgramBench asked top AI models to rebuild 200 real programs, from small tools up to things like FFmpeg and SQLite. Nine models. Zero programs fully rebuilt. Not one.

2. “Passes the tests” ≠ “good enough to ship.” The research group METR had an AI agent write 15 real code changes for open-source projects, then reviewed them by hand. None were good enough to merge as-is, not even the ones that passed the project’s own tests. METR estimated each would need about 40 minutes of human cleanup.

3. The speed boost fades; the mess stays. A study of 800 GitHub projects that adopted Cursor found output spiked in the first month, then the gains faded after two months. What stuck around: 41% more code complexity and 30% more warnings. And the piled-up complexity predicted slower work later.

Why? Because AI shines when something can automatically check whether it’s right: a compiler, a test suite, a benchmark. An image editor is judged on feel. Does the brush lag? Is the colour exactly right for print? Does that 2GB PSD from your client open without breaking a layer? No test suite catches those. So the errors pile up quietly.

A Photoshop clone is a weekend to start and a decade to finish. AI shrank the weekend. It barely touched the decade.

Code Was Never the Moat

This is the uncomfortable first principle: Adobe isn’t hard to replace because Photoshop is hard to write.

GIMP has been free for about 30 years. Free didn’t do it. Affinity is now free and excellent. That didn’t do it either.

What keeps people on Adobe:

  • The file format. Clients send PSDs. Print shops expect certain workflows.
  • Hiring. Every designer you hire already knows Photoshop.
  • The studio. Templates, plugins, ten years of muscle memory, a team that has to stay compatible.
  • The contract. Enterprises and schools buy seats in bulk, with support and legal cover.

None of that is code. So making code cheap doesn’t touch it.

The most-cited counterexample proves the point. Klarna supposedly “replaced Salesforce with AI”. Its own CEO later said they didn’t replace it with AI. The project was mostly about pulling their data into an in-house system, and he doubted other companies would copy them.

Same story with Sony: you can’t vibe-code your way out of a disc policy or a PlayStation account.

So What Is Actually Changing?

Quite a lot, just not at the top.

The cheap end of creative work is getting crushed. After image generators arrived, freelance job posts for image work dropped about 17%. Logos, placeholder art, quick commissions: anything where “good enough” is enough. In the writing and coding gigs hit by ChatGPT, the same researchers found the jobs that remained were more complex and paid more.

There’s a twist here. On one Chinese art-commission platform, when an image model leaked, prices fell 64% but orders more than doubled, and total revenue rose. Cheaper art meant way more people buying art. Disruption isn’t always shrinkage.

Juniors are getting squeezed, seniors aren’t. Stanford’s “Canaries in the Coal Mine” payroll study found employment for software developers aged 22 to 25 fell nearly 20% from its late-2022 peak. Older developers? Still growing. Meanwhile software job postings are up almost 15% since early 2025, and 71% of the past year’s increase is senior roles. (The Stanford authors are careful: their numbers describe what happened, not proof that AI caused it.)

That’s the real long-term risk. Not mass unemployment, but a broken ladder. If nobody hires juniors, where do the seniors come from in ten years?

Small, simple software is exposed. The stuff companies say they’re most likely to replace with their own AI-built tools is internal dashboards, admin panels, forms and simple workflow automation. The $15/month tool that does one thing is in trouble. The system a whole company runs on isn’t.

Pricing is where incumbents actually bleed. If AI agents do work humans used to do, companies need fewer seats. An index of software stocks fell about 26% in the first quarter of 2026, with some names down as much as 58%, while revenues kept growing. Adobe’s stock fell roughly 40% from its 52-week high despite record sales. Investors aren’t seeing customers leave; they’re betting on future pricing pressure. Expect a lot more “pay per AI action” and a lot less “pay per person”.

Who Wins

Not the angry hobbyist with a weekend and a prompt. At least not against Photoshop.

The players actually taking ground from Adobe are other big companies giving professional tools away: Canva with Affinity, Blackmagic with DaVinci Resolve. Free and polished beats free and half-finished, and it’s backed by a company that will still be maintaining it in five years.

Vibe coding does win in the places that matter to you:

  • Your own tools. The internal app nobody would ever sell you, built in an afternoon.
  • The long tail. Niche software for twelve people that never made economic sense before.
  • Prototypes. Proving an idea works before anyone spends real money on it.

That’s a real revolution. It just happens underneath the incumbents, not on top of them.

What Would Prove Me Wrong

I’d change my mind if:

  • Adobe starts disclosing that its paid subscriber count is falling. Right now it doesn’t publish that number at all, which conveniently hides any churn behind price rises.
  • A vibe-coded creative tool shows real professional usage. That means active users at studios, not GitHub stars.
  • Sony’s next results show the disc boycott biting.

Until then, the evidence says: AI made starting software almost free, and finishing it about as expensive as ever. Anyone predicting the death of the incumbents is mixing those two up.

The revolution is real. It’s just happening at the bottom of the market first.