Techies have a peculiar favourite sport: they love automating everything away – ideally themselves. Anyone who has followed the developments of the past few months already suspects that they are well on their way to taking gold in this discipline.
In the public eye, the dominant image of AI is usually that of a creative artist, copywriter or chatbot. But while the public is still debating whether AI can write real poetry, the backbone of our digital society is quietly being reshaped in the background: software development.
Software has long been the bottleneck of digitalisation. It is expensive and has so far depended on a scarce resource: developers. Worldwide, there is a structural shortage of around 4 million skilled professionals in this field. That is the problem many companies are facing.
Over the past few months, we have watched the solution emerge at record speed.
Looking back: from tool to digital team
I have been following this development for a while now. In the post “Claude 4.5 Sonnet: The first AI developer that can code for 30 hours straight” I described what I call the code paradox. Why does AI find code so much easier than human language? The answer lies in verifiability. Code either works or it doesn't: binary feedback that AI models learn from at breakneck speed. What we are witnessing right now is the leap from a mere tool to autonomous “Agentic AI” that plans, tests and iterates on its own for hours at a time.
What this feels like in real life is something I demonstrated in “From idea to app without writing a single line of code” through a self-experiment. To support my teaching assignments, I built a custom assessment application – as project lead, not as a coder. Through pure “vibe coding”, meaning describing workflows and goals in natural language, I steered AI agents that did the actual development work. In about three days, I had an application that would once have taken a whole team more than a week. I went from process manager to creative conductor of a digital orchestra.
But this personal microcosm is merely a reflection of a macroeconomic upheaval.
The hard data: coding is taking over the market
What started as an exciting POC experiment and was dismissed by the pros as "vibe coding" has, in 2026, undeniably become the killer app of artificial intelligence. Look at the current market data and the trend becomes almost physically tangible:
- The token economy: Around 50% of all LLM tokens consumed globally are now used for programming tasks. Coding isn't just any use case – it is the dominant application, the one making the server farms sweat. - In other words: AI is no longer primarily used for creative entertainment but to boost production. This isn't a consumer wave, it's a wave of industrialisation!
- The benchmark frenzy: The industry standard SWE-Bench (which tests how well AIs solve real GitHub issues from open-source projects) jumped from 48% to an incredible 93.9% within 12 months. Fact: AI agents are more efficient than a broad majority of developers. They are no longer assistants; they are production machines.
- The reality in Big Tech: We are no longer talking about prototypes. Microsoft, Google and Meta already have 20 to over 30% of their production code generated by AI. Across industries worldwide, we have already reached 41% AI-generated code.
- Startup miracle: Platforms like Cursor reached 2 billion dollars in annual recurring revenue (ARR) within nine months – the fastest SaaS growth in history, driven purely by developer adoption.
Scenarios and approaches: what does this mean for us?
When the cost of creating code is pushed close to zero, the rules of digital transformation change fundamentally. Here are three scenarios companies and project managers need to prepare for:
Scenario 1: The “build vs. buy” calculation shifts dramatically
If smart teams (as in my mini-MBA experiment) can build tailor-made tools in days rather than months, the old philosophy of almost always falling back on standard SaaS solutions collapses. Going forward, companies have a real option to run hyper-individualised software ecosystems, because source code is no longer painstakingly written by hand but generated and maintained automatically.
Scenario 2: The Trojan horse of AI adoption
Grandiose AI initiatives in companies often fail because of data protection concerns, lack of trust, or because the financial return (ROI) is unclear – an AI project in marketing is still hard to measure in hard numbers. This is where AI coding serves as the perfect “Trojan horse”: it delivers measurable, irrefutable facts. When developers suddenly save 46% of their time on routine tasks thanks to AI, that's a business case you can walk any CFO through instantly and painlessly. Once management has grown used to the safe and profitable use of AI in the IT department, the barriers to deploying AI in the rest of the company melt away almost by themselves.
Scenario 3: The end of the “developer barrier”
The fact that, at startup incubators like Y Combinator, 25% of startups in a recent batch launched with 95% AI-generated code points to a profound democratisation. The bottleneck of missing developers is widening. Programmers will shift from “code writers” to strategic architects, and business departments will be able to orchestrate digital solutions themselves, as long as they can describe the problem precisely enough.
But: It’s always the Process, Stupid!
Which brings me back to the crucial point I never tire of hammering home “It’s Always the Process, Stupid!”: There is really no such thing as an “AI strategy”; there is only business process optimisation.
Over the past decades, many bad, inefficient workflows were caused by bad standard software that we had to buy for cost reasons. Because custom development was barely affordable, we bought huge off-the-shelf applications and forced our teams to squeeze their human, organic ways of working into the rigid corsets of these tools.
Anyone who now believes you can simply upgrade rigid standard software with a bit of “Copilot AI magic” to heal broken workflows is kidding themselves: a bad system will keep dictating a bad process, artificial intelligence or not. If you automate a bad decision, you simply make bad decisions at the speed of light. Conventional AI tools bolted onto rigid standard software therefore often miss the actual goal.
But this is exactly where AI coding becomes a genuine game changer and reverses the trend.
Because once building software is no longer the expensive, limiting factor, we can put the focus back where it belongs: back at the whiteboard. We can put the problem and the process 100% at the centre, strip it down and optimise it. And only as the very last step do we instruct the AI agents to build us exactly the software that adapts to our perfect process, rather than the other way round.
We are moving from a world in which execution (the “how”) was the limiting factor to a world in which the idea and the understanding of the problem (the “what” and the “why”) take centre stage. From now on, code is merely the fleeting by-product of a well-designed solution.
Technologies change at breakneck speed.
The rules of economic efficiency don't.
