If AI can clone any app in a weekend, what is a software company worth?

THE ADOBE CLONE. In early October, US developer Brandon Thomas released ArtCraft. It is a set of seven free, open-source apps that mimic Photoshop, Illustrator, Premiere, Lightroom, Acrobat, After Effects and InDesign. According to HotHardware, he built them with Claude Opus 5.5, Anthropic’s model, and the Rust programming language. They run on macOS, Windows and Linux. Thomas says he was fed up with Adobe’s “cancellation fee”, which he calls a “dark pattern”: a design meant to trap customers. It is the most visible case of what we might call “Spite SaaS”: cloning a paid tool out of spite and giving it away.

Some perspective is needed. The apps are in alpha, which means very early builds. They have bugs, behave unpredictably and run slowly. Thomas himself admits they are not ready for professional work. But the signal matters more than the product. Marek Kowalkiewicz of Queensland University of Technology (QUT) put it this way to ABC News: copying a program used to mean hiring a hundred developers, and now it is “not expensive anymore”. That same month, a tool called Reverse Engineer Anything, which uses AI to take apart and replicate features of other software, topped GitHub’s trending list.

Our thesis is this. AI does not make software worthless. It makes code worth less. Value moves to what code does not contain: data, trust, distribution, regulatory compliance, integration and support. Companies that charged for an interface will suffer. Companies that charge for solving a whole problem will suffer far less.

WHAT CAN BE COPIED AND WHAT CANNOT. An AI model is good at copying what it can see: screens, menus, keyboard shortcuts and basic features. According to HotHardware, early PhotoCraft users say its layout closely mirrors Photoshop’s. That is the copyable part.

The hard parts sit off-screen, and there are five of them. The first is data: a CRM (the software that manages customer relationships) is valuable because of years of stored history, not because of its forms. The second is network effects: Slack or Figma are worth more the more people work inside them at once. The third is distribution: Microsoft does not sell Excel on its own, it bundles it into a contract with everything else. The fourth is compliance, which covers security certifications, audits and data protection. A hospital or a bank cannot install an alpha-stage clone downloaded from GitHub. The fifth is integration and support: someone who answers at 3 a.m. and takes responsibility when something breaks.

An analysis by 24/7 Wall St. on software stocks points the same way. Incumbent vendors can sell AI inside their existing contracts because customer data, permissions and audit history already live in their systems. Our reading: the moat has not vanished, it has moved. It used to be “this is hard to build”. Now it is “this is hard to replace without breaking the business”.

WHAT THE MARKET SAYS. Markets have already priced in the fear. In February, Bloomberg described the software selloff as the biggest one it had seen driven by fear of AI displacement. According to Forbes, citing Bloomberg, a Jefferies trader dubbed it the “SaaSpocalypse”. SaaS (software as a service) is software rented by subscription over the cloud. According to Bloomberg and Forbes, the trigger was a set of Anthropic plug-ins for its Claude Cowork tool. Thomson Reuters had its worst day on record, and monday.com fell 21%. By various financial press estimates, IGV, the iShares exchange-traded fund that tracks software companies, was down roughly 30% for the year by April.

Then came the rebound. According to 24/7 Wall St., IGV rose 41% between April 13 and October 12, twice the gain of the Nasdaq-100 (21%). The same analysis warns that the rally mostly reflects investors paying more for the same earnings, not earnings growing faster. Our reading: the market went from panic to relief without stopping to think. Software did not die in April, and it is not safe in October.

PER-SEAT PRICING IS CRACKING, SLOWLY. The deeper fear is not the free clone. It is the pricing model. Most business software is sold “per seat”, meaning per licensed employee. If an AI agent does the work of ten people, a company needs fewer licences.

That is why some vendors now charge per outcome. According to its pricing page, Intercom charges $0.99 for each conversation its Fin agent resolves without human help. Industry reports say Zendesk charges $1.50 per automated resolution. But real spending data calls for calm. Ramp is a company that handles payments for its customers and analyses their invoices. In April it reported that, across more than 200 vendors, seat-based contracts still make up 65% to 75% of spend. Usage-based pricing sits at 4% to 6%, and those shares have barely moved in a year. According to Ramp, 99% of what customers pay Adobe is still billed per seat.

Our reading: the shift is real, but it will take years, not quarters. The most likely result is a hybrid model, with a base fee plus a charge per use or per outcome.

LESSONS FROM NAPSTER, LINUX AND NEWSPAPERS. This is not the first time something stopped being scarce. Napster appeared in 1999. Over the next fifteen years, US recorded-music revenue fell by more than half, according to the industry body RIAA. The industry did not die. It recovered through streaming, which sells convenience and catalogue, not files. Open source made the Linux operating system free. Even so, in 2019 IBM paid $34 billion for Red Hat, a company built on Linux. IBM was not paying for code. It was paying for support, certification and guarantees. The internet wiped out newspapers’ classified ads, and there the damage was lasting for many of them.

The lesson has two parts. When the product becomes free, value moves to service, trust and convenience. But not every market leader survives that shift. The ones that lead it do.

THE LEGAL GREY ZONE. Copying someone else’s source code is illegal. Recreating a program without seeing its code is different. The classic technique is called a “clean room”. One team studies the product and writes a specification. Another team, which has never seen the original, builds from that spec. That is how Phoenix cloned IBM’s BIOS in the 1980s (the BIOS is the basic program that starts a computer). AI shrinks that process from months to minutes. In April, according to Slashdot, two researchers half-jokingly presented malus.sh, a service promising to recreate open-source projects with “no copyleft” licences. Copyleft is the clause that requires derivative versions of a program to stay open.

Australian lawyer Malcolm Burrows told ABC that a clone is illegal if it is a “substantial reproduction” of the original. But he added that this is hard to prove when the AI never accessed the code, and that the area will become “very grey”. There are other fronts. HotHardware notes that Adobe holds thousands of patents on features and workflows. SaaS contracts also often ban reverse engineering and using the service to build a substitute. And lawsuits have started. In July, Runlayer sued Rippling. According to the complaint, Rippling used a collaboration to build “essentially a clone” of Runlayer’s product. These are unproven allegations. In October, OpenEvidence and Doximity settled their dispute over the alleged reverse engineering of a medical AI.

There is also an irony here. A model trained on the original code can hardly claim it worked in a clean room. We expect more lawsuits and, as Kowalkiewicz predicts, more lobbying by big vendors for stronger legal protection.

WHAT IT MEANS. In the short term, the problems are real. Aaron Snoswell, a QUT researcher, warns in ABC that an indie studio that spent years on a game could see it copied in a weekend. The most exposed are not Adobe or Microsoft. They are small creators without lawyers or a strong brand. Expect cuts at companies that charged a lot for little, and pressure on jobs in teams that only maintained interfaces.

But programming is not dead. What matters has changed. Writing code is getting cheaper. Understanding the problem, guaranteeing security and taking responsibility for the result are getting more valuable. It is the same thesis we hold on jobs: AI does not take your job, it changes which work matters.

In the long run, we are optimistic. If basic software becomes almost free, a rural hospital, a charity or a freelancer will get tools they cannot afford today. That is abundance in the literal sense. Companies will have to compete on better service, not on locking customers in. One developer’s spite will not bring down Adobe. But the warning to the whole industry is clear: if your only moat was how expensive your code was to write, you no longer have a moat.