Is software becoming free?

People are now using AI to recreate advanced software like Photoshop, combine video games and build their own alternatives to services they used to pay for. Some are even building whole software suites that challenge established companies like Adobe. At the same time, ordinary people can make websites, apps, databases, customer management tools (CRM) and their own programs just by talking to AI in their own language. Things are moving fast, and what was very difficult only a short time ago is getting easier. What happens to the software industry when people and businesses can build what they need themselves?

In brief

• Ordinary people can now build their own software with AI. You can explain what you want in Norwegian and let AI write, test and improve the code. You do not necessarily need formal training in programming.
• The software industry is already feeling the change. In a global McKinsey survey, 32 percent of respondents say their organization decided against at least one software purchase because it could build the features itself with AI.
• The pace of change may matter most. Programs that would have taken a lot of development work only a short time ago can now be built and improved with AI agents. Many of these tools are still unfinished, but both the software and the AI tools building it keep improving.

1. What is actually happening?

Just a few years ago, building software was something most people left to professional developers. You either had to know how to code, learn how to do it or pay someone else.

That is no longer as necessary as it used to be. With ChatGPT, Codex, Claude Code, Grok Bot and other AI tools, you can describe what you want to build in everyday language. AI can create files, write code, set up databases, test features, find bugs and change the program as you explain what you want.

For example, you could say: "Make a program that keeps track of all my customers, their invoices and any missing payments. I want a simple overview where I can search, filter and export information."

An AI coding agent can then start building it. Once you see the result, you can carry on: "I want a different menu. Add automatic reminders for unpaid invoices. Make it work on my phone too." AI changes the program, and you test it again.

This is a very different way of working from traditional programming. You decide what the program should do, while AI handles much of the technical work. That does not mean everything works perfectly on the first try. But it does mean more people can build something useful without having to learn a programming language first.

From chatbot to AI agent

A regular chatbot mainly gives you answers. An AI agent can also take action. It can use tools, look through files, write code, run tests and keep working based on the results.

An AI model does not automatically have access to programs, files and other tools. To carry out practical tasks, it needs a system around it that provides tools, relevant data, instructions and checks. This system is often called an agent harness, or just a harness.

Examples include OpenAI Codex, which can be used through ChatGPT, Claude Code from Anthropic and Grok Bot from SpaceXAI. These are different agent products, but they share the same basic idea. AI gets tools and a place to work so it can take several steps, check the results and keep going.

Think of the AI model as the brain, and the harness as the toolbox, workspace and routines that let it get things done. An agent might build a program, run it, notice something is wrong, read the error message, fix the code and try again.

This way of working is a big reason AI can now do things that used to be much harder. And that brings us to something that has attracted a lot of attention lately: reverse engineering.

2. When AI can examine and recreate existing software

Reverse engineering means studying a finished product to understand how it works.

In traditional software development, you start with the source code: the instructions developers write to create a program. Reverse engineering starts at the other end. You have the finished program and try to work out how it was built, what features it has and how they work.

People have done this for many years, but it has often taken a lot of technical knowledge and time. AI agents can now do much of the analysis.

REA: Reverse Engineer Anything

One project that has attracted a lot of attention is REA, Reverse Engineer Anything. It is hosted on GitHub, a website where developers can store code, track changes and work together on software. Each project usually has its own code collection, called a repository, or repo for short.

When a project is public on GitHub, other people can look at what has been done. If the license allows it, they can also build on the code and contribute improvements.

REA lets AI agents connect to tools that analyze software, even without access to the original source code. The agent can study how a program behaves and how it is put together, explain how particular features work and use that knowledge to build new tools.

REA can use established analysis tools such as Ghidra. It brings existing technology together in a workflow AI agents can use. For example, you can ask an agent to investigate how a particular feature works and then build a similar feature in your own program.

That does not mean REA can automatically recover the full source code of any program. But it does make many tasks that used to be difficult much more accessible.

How good have AI models become at reverse engineering?

In August 2026, researchers published SRE-Bench, a test designed to measure how well AI agents can understand programs without their source code. The researchers created 19 private programs, 262 different binary versions and 1,572 test tasks. They wanted to find out whether AI could analyze software it had not seen before.

When OpenAI later tested GPT-6 Astra, the company reported that it solved 88 percent of the tasks on the first try and 99.2 percent within four tries. In OpenAI's comparison, the earlier GPT-5.6 Sol model scored 55.9 and 68.7 percent. These are the results OpenAI published, and the researchers behind SRE-Bench have also discussed the large improvement.

The figures do not mean AI can recreate 99.2 percent of all software in the world. The test covers specific analysis tasks. But the results show how quickly software analysis can improve from one model generation to the next.

Is it legal to recreate software?

There are legal forms of reverse engineering and independently recreating what a program does. European law distinguishes between a program's functionality and its protected code. The Court of Justice of the European Union's ruling in SAS Institute v World Programming is important here. Norway's Copyright Act also allows lawful users to examine how a program works under certain conditions.

One approach is called clean-room engineering. You document what the program does, then build an independent version from that description. But this does not mean you can freely copy the original code, graphics, music, protected files or other material covered by rights.

It is also important to distinguish between open source and an independently built copy. Open-source software makes its code available under a license that grants certain rights to use, study, change and share it.

A commercial program does not automatically become open source just because someone recreates its features. But it can get a new competitor that is.

3. From Adobe to video games: what people have already made

One of the most talked-about examples in October 2026 is a collection of programs that aim to recreate features from Adobe Creative Cloud and other commercial software.

A free competitor to Adobe Creative Cloud

Developer Brandon Thomas started ArtCraft, a set of open-source programs written in Rust and developed with extensive use of AI. Thomas has said that frustration with Adobe's subscription and cancellation practices was part of his motivation.

ArtCraft now includes 13 different programs. They cover image editing, video, vector graphics, documents, presentations, spreadsheets and technical drawing.

See all 13 ArtCraft programs

The ArtCraft programs, as of October 2026:

• PhotoCraft: Image editing, layers and retouching. An alternative inspired by Photoshop.
• VectorCraft: Vector graphics and illustration, for tasks similar to those handled by Illustrator.
• FilmCraft: Video editing, inspired by programs like Premiere Pro.
• LightCraft: Organizing and editing photographs and raw files, similar to Lightroom.
• PdfCraft: Working with PDF documents, as in Acrobat.
• EffectCraft: Motion graphics and visual effects, inspired by After Effects.
• DesignCraft: Page layout and publishing, as in InDesign.
• FigCraft: User interface design and prototypes.
• WordCraft: Word processing, similar to Microsoft Word.
• SoundCraft: Audio editing and mixing.
• DeckCraft: Presentations, similar to PowerPoint.
• CADCraft: Technical design and drawing.
• GridCraft: Spreadsheets and data analysis, similar to Excel.

You can find all the programs and their development status on ArtCraft's official website.

These are not finished replacements for all the original programs. PhotoCraft is still in early alpha. Several of the others are under development. There are bugs, missing features and limitations. A hands-on review by Creative Bloq on October 13, 2026 found problems with image processing, color management and video editing. The developer is open about this too. PhotoCraft's own GitHub page describes it as unfinished and not ready to replace Photoshop in everyday professional work.

But this is where the pace of development gets interesting. The projects are open. Other developers can study the code, report bugs and help improve it. PhotoCraft has already received a large number of proposed changes and has seen a lot of activity on GitHub. Those developers can use AI agents to build, test and improve the software. And the AI tools themselves keep getting better.

The question is not just how well PhotoCraft works today. It is how well it might work in three months, six months or a year. We do not know. But judging its future only by today's bugs gives us a very limited picture.

Frustration with QuickTime led to a new program

A smaller and perhaps more relatable example comes from Mark Kashef. He used QuickTime to review videos but wanted more control over playback speed. Instead of looking for another commercial program, he used an AI agent to build his own. He called it Tempo.

According to his own account on LinkedIn, the first working version took about 30 minutes and cost around four dollars in AI usage. Tempo had the features he needed, including custom playback speed, pitch adjustment and audio waveforms. This is his own account, not an independent quality test.

The point is simple: He did not need to build a complete competitor to QuickTime. He only needed a program that did what he wanted.

When Mario moves into Elden Ring

Video games show the same possibilities in a much more visual way. Developers are combining games, engines and game mechanics in ways that used to require a lot of specialist knowledge.

There are projects where Super Mario explores Elden Ring using the movements from Super Mario 64, and where Spider-Man's web-swinging works in Gotham from Batman: Arkham Knight. Others combine Minecraft, Grand Theft Auto, World of Warcraft, Skate 3 and Call of Duty.

Some projects recreate features. Others connect two existing games, while some build on earlier mods. Not all of them were made with AI, and many build on years of earlier work. But AI makes it easier to analyze, combine and develop these projects further.

See more examples of game mashups

Examples of game mashups:

• Super Mario 64 in Elden Ring: Mario's movements and mechanics in Elden Ring's world.
• Spider-Man in Batman: Arkham Knight: Spider-Man's web-swinging in Gotham.
• Modern Warfare 2 + Skate 3 + Minecraft: Shooting, skating and Minecraft environments combined.
• World of Warcraft + Skate 3: Skateboarding in World of Warcraft, called World of Skatecraft.
• Minecraft in Skyrim: Minecraft features in Skyrim's world.
• Minecraft in Elden Ring: Building and other Minecraft mechanics in Elden Ring.
• Minecraft in GTA V: Experiments with Minecraft mechanics in Grand Theft Auto.
• Super Mario 64 in Minecraft: Play as Mario with his original movements.
• Super Mario 64 in Garry's Mod: Mario as a playable character in Garry's Mod.
• Super Mario 64 in Sonic Generations: Mario with his own movement system in Sonic's world.
• Skate 3 in Garry's Mod: Skateboard physics and tricks from Skate 3.
• Minecraft in Outer Wilds: Minecraft mechanics in Outer Wilds' planetary system.
• Doom in Factorio: The classic Doom running inside Factorio.
• Doom in Persona 5 Royal: Play Doom inside another game.

A larger collection is available in the GitHub project Awesome Game Mashups. It lists more than a hundred projects, but distinguishes between available tools, experiments and unverified sightings.

Another interesting example is the work on the Nintendo 64 game Snowboard Kids. In August 2026, developer Chris Lewis described how he rebuilt the game's code in 84 days, with AI as part of the process.

The point is not that all of this used to be impossible. It is that tools once limited to people with a lot of technical knowledge are now becoming available to far more people. And games show how much more becomes possible when you are no longer limited to what the original developer chose to make.

4. You do not need to be a programmer to build software

I have seen for myself how big this change is. My background is in photography, Photoshop and visual production. I am not a trained software developer or accountant. Yet I now build websites, databases, programs and other systems from scratch with AI.

One of my projects is an AI-based accounting system for my own sole proprietorship. I wanted something that fitted what I actually need. I do not run a company with lots of employees, payroll and complicated processes. So I do not need every feature a large commercial accounting system offers. Instead, I have built my own with AI. When I need a new feature, I ask AI to look at how it works in other accounting systems and read the relevant official documentation. AI then builds and tests it. I test it several times myself and point out anything that is wrong or does not work well.

The system was developed around Norwegian accounting and bookkeeping rules, using information from sources including the Norwegian Tax Administration, its GitHub account, Altinn and Lovdata. It also includes features for data storage, logging and backups.

It has not worked perfectly from day one. I have spent time building, testing, finding bugs, fixing them and building further.

That is an important point: you do not need to know everything before you start. You can learn as you build. When I come across something I do not understand, I ask AI to explain it. When I need to know more, I ask it to look into documentation, laws, rules or other relevant sources. This has let me take on projects I previously did not have the technical skills to complete.

That does not mean AI is always right. It means I can use it both to do the work and to learn what I need to check the result. This is often called human in the loop: a person still takes part in reviewing, checking and making decisions.

For me, perhaps the biggest difference is that I do not write a single line of code by hand. I mainly use ChatGPT, Codex, Claude Code and Grok Bot, and regularly try other models too. And I do most of the work simply by speaking Norwegian to AI.

What I find really cool is that I do most of the development while out walking, driving or relaxing on the sofa in the evening. That is a long way from the traditional way of building software. I explain what I want to make, discuss ideas, ask questions and tell AI to do the work. When I get a new idea, I can talk about it straight away. I do not need to sit down at a computer and start writing code.

Talking works especially well for me when I am developing ideas. The thoughts keep coming, and I can explain them, change direction and build on them as I speak. The AI agents handle the coding. I can then look at the result, test the features and explain what I want to improve.

And it is not just accounting. I also build websites and systems without relying on traditional website builders like Wix or Squarespace. That does not mean everyone should do the same. It means the option is there.

Kevin Luke built his own business system with Grok Bot

Kevin Luke is not a programmer. He runs Above the Line, a company that provides technical production and equipment for events in the US and internationally. In an interview during the Grok Bot for Sales workshop in October 2026, he explained how he uses Grok Bot to run and develop parts of the business.

He previously used several different programs, including QuickBooks for quotes and Flex for inventory. He wanted one system that brought the business's needs together. So he asked Grok Bot to build AboveHub, his own system for customers, quotes, invoices and stock. The AI agent was given access to earlier invoices, quotes and inventory records to help build the system. The result was software tailored to his business.

Luke says he can also ask the agent to make changes when employees want new features. He uses his phone, among other things, to explain what needs changing.

He also uses his own AI agents to research potential customers, find events and prepare sales proposals. One agent, Clarity, handles research. Another, Synergy, works on sales and marketing. Luke has combined software built for his business with AI agents that use it to get things done.

People still need to check important results, including quotes that will be sent to customers. But this is very different from buying a ready-made system and adapting the whole business to the way its supplier decided it should work. Businesses can now do much more to build software around the way they actually work.

5. AI can also use software for you

There is another change that matters for the future of software. AI does not necessarily have to build a new program. It can also use existing software on your behalf.

This is called computer use. AI gets access to a graphical interface, the same kind people use when working on a computer. It can see what is on the screen, move the mouse, click buttons, type, open menus and carry out tasks.

OpenAI demonstrated this with its Computer-Using Agent in January 2025. The technology has moved on since then. AI agents can now use browsers and other programs to get things done. What an agent can actually access depends on the software, permissions and environment it works in. Some agents use their own computer in the cloud. Others use a browser or a limited local environment.

The basic idea is the same. You tell AI what needs doing, and the agent takes the steps. That could mean opening a website, looking up information, filling in forms, finding documents or testing a feature in a program.

In creative software, AI can also use tools and automation to work with images, video and 3D. Programs like Blender can already be controlled through AI integrations, and several ArtCraft programs have been developed with support for agent control.

This raises an interesting question: How important will the user interface be if AI can do the work for us?

You may not need to learn where every feature in a program is hidden. You need to understand what you want to achieve. AI can help you use the tools. And it can explain how they work if you want to learn.

That does not mean every task can be handed to AI without checking. But it could change how much time we spend operating software ourselves. Perhaps the software that matters most in the future will not be the one with the most buttons and features. It will be the program or system that does the job best.

AI can also help organize your personal life

This is not just about programming, businesses and jobs. Think of all the little things we spend time on at home: money, insurance, documents, bills, subscriptions, appointments and deadlines. AI can already help with much of this. With AI agents, more of these tasks can be done automatically, provided the agents have access to the information and tools they need.

Here are some examples of what you could use AI for:

• Personal finances: Bring together bank transactions, bills and other financial information, show where your money goes, spot unnecessary expenses and suggest ways to save.
• Insurance: Gather your policies, explain what they cover, compare terms and prices from other companies and flag when a policy needs reviewing.
• Receipts and warranties: Keep proof of purchase, track warranties and deadlines for making a claim, and find the right receipt when something breaks.
• Bills and subscriptions: Keep track of what you pay for, spot services you may not use, look for cheaper options and remind you about upcoming payments.
• Your home and maintenance: Organize property documents and keep track of service agreements, warranties, electricity contracts and maintenance tasks.
• Family and everyday life: Keep track of appointments, activities, travel and important dates, and remind you when something needs attention.
• Serving on a club or housing cooperative board: Bring together bylaws, meeting minutes, contracts, accounts and other records in one system. AI can help with meeting notices, minutes, following up decisions and keeping track of deadlines.
• Selling used items: Check market prices, organize product photos, write listings and help manage the sale.

Several of these tasks can be combined in one personal system, rather than using a different app or service for each one. You can also set up routines where an AI agent checks information and only tells you when something needs attention. For example, it could review your subscriptions every month, check whether an insurance policy is due for renewal or give you an overview of important tasks for the coming week.

You can still decide which documents and accounts AI can access, and which actions need your approval.

The point is that AI no longer has to be just a tool you open when you have a question. It can also become a system that helps you stay organized, keeps an eye on what matters to you and handles tasks you used to do yourself. And just as with business software, you can build it around your own needs.

6. Today's limits are not necessarily tomorrow's limits

I think this is one of the most important points when we talk about AI. We have already seen the same pattern many times. Something works badly. People point out the faults. New models arrive. The results improve. Then the discussion moves to the next thing AI cannot do yet.

What happens when AI gets better at finding and fixing bugs?

This matters especially for software. If an AI agent builds a program with a bug, it can also be asked to investigate it. It can read error messages, run automated tests, compare what should happen with what actually happens and try different fixes. It does not necessarily need a person to write the fix. People can still set the requirements and check the work, but much of the bug fixing can be automated.

In February 2026, Anthropic described an experiment where 16 AI agents worked together to build a C compiler. A compiler is a program that turns source code into software a computer can run. In other words, it is one of the tools developers use to make other programs.

The result included around 100,000 lines of code and could compile the Linux kernel. The project took nearly 2,000 Claude Code sessions and cost about $20,000 in API usage. It shows both how much AI agents can do and that large projects can still take considerable resources.

But the cost of building software is changing. And the tools that develop, test and check it keep getting better. That is why I think it is a mistake to look at today's problems in PhotoCraft, FilmCraft or other AI-built programs as permanent limits.

What does not work today may work in the next version. There is no guarantee every problem will be solved, or how long it will take. But the progress of the past few years gives us good reasons to watch what happens next.

7. What does this mean for software companies and jobs?

Seeing people build their own software is one thing. Finding out whether this is already changing how businesses buy technology and organize their work is another. We now have some concrete figures.

32 percent have already chosen to build instead of buy

In August 2026, McKinsey published its global survey, The State of AI in 2026. It covered 1,719 respondents from 97 countries.

32 percent said their organization had decided against buying at least one software product or feature because it could build the functionality itself with AI coding agents. Among organizations with the strongest documented results from AI, the share was even higher, close to half.

That does not mean 32 percent of all businesses worldwide have stopped buying software. But it does show that the alternative we are discussing already affects buying decisions.

The choice used to be between buying a ready-made system and paying developers to build your own. There is now a third option: use AI agents to build it in-house, with much less coding by hand.

This could affect the whole SaaS market. SaaS, short for Software as a Service, means software delivered as a service, often with a monthly or yearly subscription. Customer management systems, accounting software, project management tools, document systems and website builders are common examples. If more customers can build their own tools, they have another option when deciding whether a subscription is worth paying for.

Businesses have already started changing

According to Statistics Norway, nearly five in ten Norwegian businesses with at least ten employees used AI technology in 2026, up from three in ten in 2025. Among large Norwegian businesses, the figure was about eight in ten. This covers all AI use, not just advanced agents. Even so, it shows how quickly the technology is being adopted.

Several companies have also linked major changes in their organizations directly to AI.

DNB in Norway

On October 6, 2026, DNB announced plans to cut around 400 full-time equivalent roles in Technology & Services. The bank says AI agents are already taking over tasks that used to be done by hand, including customer checks, technology development and programming. AI is an important part of the explanation, alongside other organizational and technological changes. This is a concrete Norwegian example of AI agents already affecting tasks and staffing in large businesses.

Source: DNB's own press release.

WiseTech Global

In February 2026, Australian software company WiseTech announced plans to cut around 2,000 jobs, nearly 30 percent of its workforce. The company linked the changes to extensive use of AI in programming, customer service and internal processes. CEO Zubin Appoo described this as a fundamental change in how software is developed.

WiseTech is particularly interesting because it has also started changing how it prices its own services, including a greater focus on charging per transaction.

Source: ABC News Australia.

These examples show that AI is already affecting jobs. But it is not only being used to reduce staff numbers.

Norway's sovereign wealth fund: helping employees do more

In May 2026, Nicolai Tangen, head of Norway's sovereign wealth fund, said all 680 employees used AI every day. More than half also used coding tools. Tangen described how employees who had not previously done any programming could now use AI to build their own tools, analyze data and automate tasks. Portfolio managers could analyze more information. Lawyers could handle cases faster. Other employees could build tools to help with their work.

Tangen also reported a self-assessed productivity increase of more than 20 percent. The fund's deputy head later clarified that this was not a scientifically established measure of impact, but was based on employees' own assessments.

The interesting part is still the way people work. AI gives employees access to skills and tools they did not have before. The fund has not described this as a plan for machines to replace everyone. Tangen has instead stressed that the technology makes people more productive.

Sources: Tangen's speech published by Norges Bank and NTB's report on the productivity figures.

What happens to jobs next?

I think we will see more changes like these. Some tasks will be automated. Some jobs will disappear or change. At the same time, AI will create a need for new skills, new ways of working and probably new kinds of jobs.

A global PwC report, published in June 2026 and based on more than a billion job advertisements, shows that demand for AI skills is growing faster than the wider job market. Roles that ask for AI skills also pay significantly more on average than similar roles that do not.

That does not mean everyone needs to become an AI developer. But I think knowing how to use AI will become a more important part of many jobs. Being willing to learn something new will matter too.

History is full of jobs and tasks that changed or disappeared because new technology took over. Think of the lamplighters who used to go around lighting and putting out street lamps by hand. When the technology changed, that particular task was no longer needed.

The same principle applies today, but it is no longer just physical work that can be automated. Programming, analysis, administration, customer service and other knowledge-based tasks are affected too. The difference is that the same AI technology that automates tasks can also help people learn new skills and create something new.

The same AI tools that are changing how we work also make it easier to build the software we need. That does not mean all software will become free. Ready-made products can still be worth paying for, partly because the supplier takes responsibility for running them, keeping them secure, updating them and providing support. But people and businesses now have more ways to build what they need themselves. That could change what they are willing to pay for.

And perhaps the most important question is not whether AI can already replace the program you use today. Perhaps it is whether you will still need to buy it a year from now.

Sources and further reading

This article draws on official publications, research papers, company reports, developers' own projects and published interviews. Some results and experiences are reported by the developers or companies themselves.

AI development, research and statistics

• McKinsey: The State of AI in 2026, August 25, 2026.
• Statistics Norway: AI use in business is accelerating, September 25, 2026.
• OpenAI: GPT-6 Astra, documentation and test results.
• SRE-Bench: The Next Challenge for Agentic Cybersecurity, research paper, August 2026.
• Anthropic: Building a C Compiler with Parallel Claudes, February 2026.
• PwC: 2026 Global AI Jobs Barometer, June 2026.

Software, reverse engineering and AI agents

• REA: Reverse Engineer Anything, GitHub.
• ArtCraft: All 13 programs.
• PhotoCraft: Source code and development, GitHub.
• Creative Bloq: Hands-on review of the ArtCraft programs, October 13, 2026.
• Mark Kashef: Tempo, his QuickTime alternative, October 2026.
• OpenAI: Computer-Using Agent, January 2025.
• Grok Bot for Sales: Kevin Luke and Above the Line, workshop published in October 2026.
• Court of Justice of the European Union: SAS Institute v World Programming.
• Lovdata: Norwegian Copyright Act.

Companies and jobs

• DNB: Organizational changes in Technology & Services, October 6, 2026.
• WiseTech Global: AI and planned job cuts, February 2026.
• Norges Bank: Nicolai Tangen's speech on AI at Norway's sovereign wealth fund, May 2026.
• Digi/NTB: Clarification of the fund's productivity figures, June 2026.

Games, rebuilding and mashups

• Awesome Game Mashups: Project collection.
• Chris Lewis: Decompiling a Nintendo 64 Game in 84 Days, August 2026.

More sources on AI agents

• OpenAI: Unrolling the Codex Agent Loop, January 2026. How an AI agent uses models and tools to work through a task.
• Anthropic: How Claude Code Works. An explanation of AI models, tools and the agent harness.
• Grok Bot: Official documentation. AI agents that work on their own computers and use connected services.
• GitHub: About Repositories. How GitHub is used to store, share and develop software.
• OpenAI: Using Cloud Browser in ChatGPT. How AI can use browsers and carry out tasks.
• Grok Bot: Use Cases. Examples of AI agents working with documents, email, follow-ups and regular routines.

Sources reviewed as of October 15, 2026. These software projects are being developed all the time, so features, versions and availability can change quickly.