If AI Can Copy Photoshop, Why Would Anyone Build the Next Figma?

Earlier this month, a developer named Brandon Thomas released free copies of seven Adobe apps, including Photoshop and Premiere. He says he built them with Claude Opus 5.5. If one person can copy Adobe with AI, will we get more new software from now on, or less? I can see it going both ways.

The Photoshop copy is called PhotoCraft. Its GitHub page calls it a "clean-room reimplementation of Adobe Photoshop" and says it's an early alpha that isn't ready for daily professional work yet. Even so, it has more than 72,000 stars on GitHub, and it only went up on September 30. Thomas wrote "Software is over," according to Ars Technica. On Hacker News, he said that getting to 99% of what Adobe's apps do "will take a while, but I'm sure it'll be measured in months and not years."

On the one hand, a lot of software never got built because it was too much work. You had an idea for a tool, but you needed a few developers and a year or two, so it stayed an idea. Now you can build it yourself with Claude Code or Codex and see if people want it. Finding people who pay for it is the hard part, as I wrote in Anyone Can Build an App Now, but Who Is Going to Pay for It? But many more ideas get tried now, and some of them will be things nobody has built before.

On the other hand, there's Figma. Dylan Field and Evan Wallace started it in 2012. According to Figma's IPO filing, they spent three years building and testing it with users before they launched it. The filing says it was the first design tool that ran in the browser and still worked like a desktop app. In 2022, Adobe agreed to buy Figma for about $20 billion. The deal fell apart in December 2023 because it couldn't get past regulators in Europe and the UK.

Adobe already had its own design tool back then, called Adobe XD. The Register reported that Adobe's general counsel said in 2022 that XD made no more than $15 to 17 million a year. Adobe itself said Figma was expected to end 2022 with more than $400 million a year in subscription revenue. So instead of copying Figma, Adobe offered $20 billion for it. I think that shows how hard it was to copy good software before AI, even for a company like Adobe.

Before AI, copying software was about as much work as building it in the first place. Nobody outside the company could see the source code. If you wanted your own Figma, you had to build it from scratch, and that took years. That was the protection for whoever had the idea first.

Outside of software, inventors get that protection from patents. If you invent a new machine, you can patent it, and for 20 years nobody else is allowed to build it without your permission. That's why it makes sense for a company to spend years and millions on something new. Software doesn't really have that. The European Patent Convention says that "programs for computers" are not inventions as such. There are some software patents, but it's different.

Copyright protects the code, but an AI copy doesn't need the code. PhotoCraft says it was built "from public specs and observed behaviour only." An Australian lawyer told ABC News that copyright law isn't ready for this, and that "it's going to get very grey."

Now say Figma starts today. Field and Wallace raise money and spend three years building their design tool. On launch day, everyone can see what it does. A few weeks later, someone rebuilds it with AI and puts it on GitHub for free. Would investors still pay for those three years? I'm not sure they would. And if nobody pays for the three years, the next Figma doesn't get built.

But maybe the next Figma doesn't need three years anymore. If two people can build it with AI in a few months, they don't need much money from investors. They also don't need a $20 billion sale at the end to make it worth it. Then we get a lot more new tools from people who build what they have in mind, and fewer companies that spend years on one new idea. I don't really know if that adds up to more innovation or less.

What I'm building

Delegate tasks. Get software.

Give Vroni a GitHub issue, bug report, spec, or rough idea. It reads the repo, plans the change, writes code, runs checks, and works toward a review-ready pull request.

Take a look at vroni.com