Artificial intelligence no longer just improves our work – it improves itself. This radically accelerates the speed of development and could lead to an intelligence explosion.
Feb 22, 2026, 10:24 p.mFeb 22, 2026, 10:24 p.m
Raffael Schuppisser / ch media
Since ChatGPT was launched in November 2022, artificial intelligence (AI) has catapulted itself into the collective consciousness and established itself in the everyday lives of many office workers. Answering emails, writing reports, creating illustrations and tables – chatbots such as ChatGPT, Gemini or Claude are used for all this and much more. For some it is still just a gimmick, for others it is a real gain in efficiency and for others it is impossible to imagine life without it.
When artificial intelligence becomes a model student in the workplace.image: getty images
Nevertheless, artificial intelligence has not yet led to a real revolution in the world of work and beyond. There have been layoffs in some industries that can be directly attributed to the development of artificial intelligence. But there were no job massacres and entire industries did not collapse – as was so often predicted.
So isn’t AI all that revolutionary? People still like to laugh at the curiosities that ChatGPT spews when the “brain” wired onto hundreds of thousands of chips begins to hallucinate. And some people are looking eagerly to Wall Street to see whether the AI bubble on the stock market will burst.
So is it all just hype? Two recent developments suggest that this is not the case. We probably shouldn’t feel too confident about machine intelligence. Because it could happen faster than we think.
In an essay that attracted great interest on
So what are these two developments mentioned?
First: AI agents can do much more than chatbots
While most people just chat with ChatGPT and often use free versions and are therefore still more or less at the level of November 2022, they overlook the fact that artificial intelligence can already do much more. She is no longer limited to the small input chat field, but can carry out actions independently on her own computer or on the Internet, such as rearranging photos or files.
The agent “sees” the screen, moves a virtual mouse, clicks through menus, opens a spreadsheet, copies data from emails and enters it into a form. Or he books hotels independently on the Internet. If you give him access to his credit card or his Bitcoin wallet, you can also let him pay or trade. Just about anything that runs on a screen can be taken over by AI.
The fact that companies are rather hesitant to provide information about this and that AI agents have so far been used primarily by techies is because they can make mistakes and this can be fatal if you give them access to all files or even your credit card. Of course, that doesn’t stop companies from further developing the technology and making it more secure.
It’s probably like the launch of ChatGPT: While the established tech giant Google was still hesitant to launch its AI chatbot, the newcomer Open AI rushed forward. A company will soon unleash its AI agents on the majority of Internet users.
AI agents have already become indispensable in software development, i.e. among experts. This leads to the second development, which will massively accelerate the progress of AI.
Second: The AI programs itself
When Open AI developed the first model of ChatGPT, it took over a year to produce a new version of a language model.
In 2025, several innovations appeared per year – they are now being developed almost every month. Both OpenAI with GPT 5.3 Codex and Anthropic with Claude Opus 4.6 recently presented new, powerful models almost at the same time. The acceleration in intelligence development comes about as programmers use AI to develop even better AI.
Dario Amodei, the CEO of Anthropic, says that AI now writes “a large part of the code” in his company. According to the company, ChatGPT’s new model is also the first to have been significantly involved in its own development. The now released version 5.3 is being used to develop an even better version 5.4 – and so on.
Dario Amodei, managing director and co-founder of Anthropic.Image: AP
So the skill jumps not only happen faster, they also become larger. This leads to an explosion of intelligence. Amodei is convinced that we are “only one to two years away from the current generation of AI autonomously developing the next.”
This does not necessarily lead to superintelligence or consciousness in machines – although Amodei does not rule it out. But it has significant consequences for us.
Amodei illustrates this with a thought experiment: Imagine it’s 2027. Overnight, a new country with 50 million citizens appears, and every single one of them is smarter than every Nobel Prize winner who ever lived.
You don’t want to mess with such a state, you want to cooperate with it. For the global political stage, this meant that it was crucial that the “new people” were among the “good guys”, i.e. that non-authoritarian regimes like China won the AI race. And for all of us in our everyday work?
The Anthropic boss’s forecast: AI will eliminate 50 percent of entry-level office jobs within one to five years. But as a look at his own company shows, even highly qualified jobs in the IT industry are not secure.
We should therefore be interested in working with these AI Nobel Prize winners, having them on our team and leading it.
Matt Shumar’s advice, who compares the situation to the false relaxation before Corona, was to adapt as quickly and as well as possible. Most jobs will not simply be replaced by an AI, but by a human who knows how to best use the AI.