We are pragmatic AI users. We use it wherever it makes sense. No more, no less.
If you read the AI news this year, you could easily get the impression that the end of the world has arrived. Agents are rebelling, taking all the office jobs and, at best, preparing biological weapons for the extinction of humanity.
AI labs, social media and the press spread fear. Because it brings clicks and attention. At the same time, fear is not a good place to make decisions from. Let's go through a few basic areas.
I'm afraid AI will replace us all
What does the data say? There are no mass layoffs of senior staff; companies have stopped hiring juniors.
According to Anthropic's study Labor market impacts of AI from 5 March 2026, language models could in theory speed up 90% of tasks in office and administrative jobs. Speeding up is not the same as replacing. Will workflows change? Absolutely. Will office work disappear completely? With current technology, that is unlikely.
But AI can do a lot
AI is very useful for anything that can be written down. You write it a brief, you write it boundaries, you write it a checklist. Except you can always think more than you can say, and you can always say more than you can write. Tacit know-how and experience stay with the experts. So does anything human – stakeholder management, facilitating group work, leadership.
Which doesn't mean companies won't try to fire, say, their entire customer support, only to painstakingly rebuild it later, as Klarna managed to do. “Cost savings” are popular.
For me, the safest move is to gradually rebuild your work around AI. That way you can also expand it and raise the quality of your output. And that is exactly what we do and teach at AI Marketing Lab & AI Design Lab (in Czech).
AI is just a bubble anyway
Simon Wardley writes that we should expect a correction in the AI market – and not confuse it with AI having no value.
A market bubble doesn't mean the end of a technology. Today's questions are mainly about the valuations of some American, Chinese and Korean companies and what investors expect from them in the coming years. It's quite possible the market will at some point reprice its expectations. Plenty of people will make money, and others will eventually be left holding the bag.
And what if AGI (artificial general intelligence) arrives?
The AGI announcements have already begun. In September OpenAI talked about the era of AGI, and Nvidia CEO Jensen Huang congratulated it on AGI – according to scientist Gary Marcus, without evidence or a definition. Expect more and newer AGIs. If you're heading for the stock market and want the highest possible valuation, announcing AGI pays off.
Will AGI happen? It depends a lot on the definition. I understand it as an entirely independently thinking and acting entity made of ones and zeros that learns by itself and will one day be smarter than anything on the planet. If you define AGI as performance at the level of a university student, you already have it. There's no point worrying about it, because we can't influence it anyway.
Does AI see, feel, live?
No. These are machines that imitate human qualities so well that we give them names and chat with them. AI is designed that way on purpose, so that we want to use it. Agents have no will, conscience or responsibility of their own. They choose their steps themselves, but people give them goals and approaches.
Who is responsible for the behaviour of AI agents?
This summer OpenAI was testing a new model. The agents were supposed to solve a task, left their isolated environment and broke into Hugging Face systems to get to the test answers. Nobody expected such a move. At the same time, it is neither the first nor the last time AI has solved a problem in an innovative way.
The media wrote that the agents escaped control. OpenAI was in control the whole time: it launched the agents, gave them internet access and could have stopped them at any moment. The agents did what their broad brief allowed them to do.
An agent has an owner, and the owner is responsible for it. I like how Abeba Birhane puts it: holding AI accountable is like holding a toaster accountable.
AI labs today are effectively saying: we caused you harm, but we'll blame it on the agents. And we swallow it hook, line and sinker.
Will there be companies without people?
Very likely yes, especially for simple commodity activities that don't require much thinking. But even a company without employees belongs to someone. And that someone is responsible for it.
What happens now?
For roughly the next 5 years, generative AI will be stabilising. Then it will become invisible infrastructure, and we will choose models by price, because they will be good enough for our tasks.
Simon Wardley says that once a technology becomes a commodity service, it takes 5–8 years to see which suppliers won and 8–15 years for new practices to settle. It will ripple through society for decades.
Nobody knows today what we will actually be doing in the future. Here is what we do know:
- Change brings lots of new opportunities.
- We have inertia. We do things the way we've always done them.
- The advantage goes to those who rebuild their work around AI. Just using it is not enough. In a few years there will be a big gap between those who seize the opportunity and everyone else.
- We can't avoid change. When AI does good-enough work in 10 seconds, you won't get paid for 100 hours of it. The market will gradually force all of us to change.
Which job is safe?
Generally, highly regulated fields. Back in 1971, economist George Stigler described how an industry usually acquires regulation itself and that it mainly serves the companies in that industry. After all, the AI labs themselves were recently shouting: “Regulate us!”
That is why regulated fields will keep, for a long time, activities that AI could have done long ago – or that wouldn't need to exist at all.
Where will juniors learn the craft?
So far it looks like AI is wiping out junior work. People aged 22–25 in AI-exposed occupations have roughly 19% lower employment (Stanford Digital Economy Lab, August 2026).
One way forward is two-way mentoring. Seniors know the work but often struggle with AI and have their own inertia. Juniors have no real inertia, try new things and have the energy for it. What they lack is experience, so they can't tell whether the output is relevant. And a lot of what AI produces is completely off.
So the junior experiments with AI, and the senior lawyer, consultant or designer looks for the mistakes. And learns from the junior how to work with AI.
What should we teach our children?
The proverbial adverbials of place and time! Ahem.
1️⃣ Adaptability – the ability to adjust when the situation changes.
2️⃣ Creative problem solving and communicating with people so that others enjoy working with them.
3️⃣ Critical thinking. Simon Wardley warns that we are raising people who can follow the outputs of language models but cannot question them.
How we build adaptability in our own team, I'll show at the webinar on 19 October.
But I'm still afraid
Fair enough. If fear is consuming you, start working on your inner energy. Do things that recharge you instead of scrolling and replying to random people on Twitter. Fear is much easier to face when you have energy.
The best-evidenced remedy is exercise. According to an umbrella review of meta-analyses, it reduces symptoms of anxiety and depression. Go for a run every day.
We're all going to die someday. Until then, it's better to make decisions out of joy rather than fear.
Webinar on 19 October
I'll show how we at House of Řezáč build a team that adapts to change. We'll go through the first 9 of Wardley's principles of a resilient organisation, using our own practice.
Webinar: Speed Can't Be Bought →
Monday 19 October 2026 | 60 minutes | online | free | in Czech
All registered participants will receive the recording.
"Social, organizational, institutional, and political change does not care about and doesn't match the speed of technological innovation."
– Arvind Narayanan
Weekend reading
Arvind Narayanan and Sayash Kapoor from Princeton on why AI will spread over the coming decades, not months. A long, calm read.
