One developer is building free alternatives to Photoshop and Microsoft Office, while another used hundreds of billions of AI tokens to reconstruct a blockbuster video game. The consequences for the software industry could be enormous.
For decades, companies like Adobe and Microsoft enjoyed an advantage smaller competitors struggled to overcome: their software was extraordinarily difficult and expensive to replicate.
Building something comparable to Photoshop, Illustrator, or Excel required years of development, specialized engineers, and ample financial backing.
Artificial intelligence is beginning to challenge that assumption.
A new generation of developers is using AI coding agents to create increasingly sophisticated applications, sometimes with teams so small they would barely qualify as startups.
And two developments in October suggest the technology is advancing faster than many established software companies might prefer.
One developer takes on Adobe and Microsoft
Software engineer Brandon Thomas has attracted attention for ArtCraft, an open-source collection of creative applications built with assistance from Anthropic’s Claude AI.
The project includes alternatives to Adobe products such as Photoshop, Illustrator, Premiere Pro and Lightroom.
Thomas has now expanded into Microsoft’s territory, introducing WordCraft, GridCraft and DeckCraft as alternatives to Word, Excel and PowerPoint.
The applications are free, and their development shows a growing movement known as “vibe coding,” in which developers describe what they want while AI tools produce much of the underlying software.
For consumers frustrated by expensive subscriptions, the appeal is obvious.
But there is an important catch.
Independent testing by Creative Bloq found that the ArtCraft applications still fall far short of Adobe’s professional tools. Problems include inconsistent image editing, limited compatibility, and missing advanced features.
The software is still in early development.
It would therefore be premature to declare Photoshop or Microsoft Office obsolete.
What matters is how quickly a small operation has created recognizable alternatives to products backed by enormous engineering organizations.
A video game experiment reveals AI’s growing power
An even more ambitious project emerged this month.
Developer Maurice Heumann revealed that he had spent approximately three months directing AI agents to reconstruct the software behind 2009’s Call of Duty: Modern Warfare 2.
He estimates the effort consumed between 600 billion and 700 billion AI tokens, the units used to measure information processed by language models.
The project eventually involved more than a dozen agents working simultaneously.
According to Heumann, his team reconstructed 99% of the game’s functions, and 83% matched the original, as his team verified.
The reconstructed code remains private, and the results are based on the developer’s account rather than a publicly released independent audit.
Still, the experiment illustrates an important point.
AI systems can now tackle substantial programming projects for extended periods, provided developers establish effective methods to coordinate and verify their work.
Heumann also discovered the limitation.
Early versions appeared working but contained serious underlying errors. Progress improved only after he introduced strict automated verification.
Generating code was relatively easy. Proving that it worked correctly was considerably harder.
The subscription business faces a new calculation
That distinction matters for Adobe, Microsoft, and countless companies that depend on recurring software subscriptions.
Businesses have traditionally purchased large software packages even when employees use only a fraction of the available features.
AI-generated applications could eventually offer another option.
Instead of paying for a complicated platform, a company might commission a small internal application designed around its actual workflow.
The economics are shifting from paying for access to finished software to paying for the computing power, development, and maintenance needed to create customized tools.
However, AI tokens are not free, and software still needs testing, security updates, and long-term support.
A quickly assembled application can become expensive if it breaks, exposes sensitive information, or requires constant repair.
Microsoft is preparing for a different future
The established companies are hardly ignoring the challenge.
Adobe has been expanding AI assistants across Creative Cloud, permitting users to describe tasks that its applications then perform.
Microsoft is pursuing a wider platform strategy.
In a recent GeekWire interview, GitHub Chief Operating Officer Kyle Daigle discussed Microsoft’s efforts to connect coding tools, AI agents and cloud infrastructure.
The strategy suggests Microsoft sees opportunities beyond selling individual applications.
Even if customers build more of their own software, they may still depend on Microsoft’s development platforms and cloud services.
Meanwhile, French AI company Mistral has introduced a public preview of Mistral Large 4, nicknamed “Le Chonk,” with its open model weights expected later in October.
The expansion of powerful AI models could give developers additional choices.
The hidden cost of letting AI build everything
Security may ultimately determine how far the revolution goes.
On October 6, software infrastructure company Temporal announced that much of the team from authorization specialist Oso was joining its organization to strengthen security controls for AI agents. Tiny separately acquired Oso’s software business.
The arrangement shows a concern circulating throughout the industry: Autonomous systems need strict limits on what they can access and change.
A developer who can create an application through natural-language instructions may not understand every vulnerability in its code.
That could become a serious liability when financial records, customer information or critical business systems are involved.
The real threat to traditional software giants is not that AI has already perfected free alternatives to their products. It is that producing credible competitors may become dramatically easier.
Adobe and Microsoft still possess major advantages in reliability, compatibility and buyer trust.
But if AI continues reducing the cost of building software, those companies may eventually face a difficult question.
Why should customers keep paying for complicated applications when they can increasingly build the specific tools they need?
