For five minutes, the whole of Sweden stood still.
At 4:50 on the morning of Sunday, 3 September 1967, every car in the country had to stop. Drivers crossed slowly to the other side of the road and waited. At 5:00 the radio gave the signal. A country that had always driven on the left pulled away on the right.
The photographs from that morning are famous. Stockholm gridlocked, people standing on car roofs, chaos that looks like the end of the world in 4:3. It's the most repeated part of the story — and the least important.
Twelve years earlier, Swedes had voted on it in a referendum. Eighty-three percent voted to stay on the left. They didn't want the change.
The Riksdag decided to do it anyway, on 10 May 1963. And instead of setting a date and hoping for the best, it gave itself four years. Four years for something that would happen in five minutes.
Those four years ran on two tracks at once.
The first track was concrete and sheet metal. Around 350,000 road signs to remove or replace, some 20,000 in Stockholm alone. Junctions rebuilt where a left turn suddenly became a right turn. Buses with doors on the wrong side — rebuilt or replaced. Traffic lights. Bus stops. Everything physical.
The second track was human habit. An information campaign that got in everywhere: television, radio, a song contest about the switch, the logo on milk cartons and on underwear. Not because someone in Stockholm had a sense of humour. Because the reflex to turn left sits deeper in a driver than any regulation, and you don't change a reflex with a memo.
Neither track would have worked alone. Perfect signage with drivers running last week's reflexes is a massacre. Perfectly briefed drivers at junctions built the other way round — exactly the same.
And now the part nobody quotes.
In the first weeks after the switch there were fewer accidents than before it. Not more — fewer. On the Monday after the changeover 125 collisions were reported, against 130–198 on previous Mondays. Motor insurance claims fell by forty percent.
The reason is mundane, and it's the most interesting thing in the whole story: everyone was paying attention. People drove slower, looked twice, didn't trust their own reflexes.
And then the attention ran out. Claims were back to normal within six weeks. By 1969 the accident figures were back where they had been before the change.
The infrastructure change stuck. The alertness didn't. Alertness isn't a state you can live in. It's a response to an event, and events end by definition.
Your company is changing which side of the road it drives on.
Nobody calls it that. They call it an AI rollout, a transformation, or just “we have to do something about this”. But the mechanism is identical: a large number of people have to start doing something daily that they currently do, by reflex, differently. And they have to do it in an environment that isn't set up for it yet.
The sentences I hear on first calls are almost identical from company to company:
„We bought licences for everyone. Fifteen people log in.”
— HR Director, manufacturing
„The board is asking about ROI on AI. I don't know where to start counting.”
— CFO, logistics
„We ran a training in March. It's September and nobody remembers.”
— L&D Manager, FMCG
„IT says it's a compliance risk. The business says competitors already have it.”
— COO, retail
„Three teams use three different tools. Nobody agreed on any of it.”
— IT Manager
„People are afraid to admit they use AI, because they don't know if they're allowed.”
— HR Business Partner
Notice that none of these are technology problems. Not one. They're organizational problems dressed as technology problems — which is exactly why buying licences doesn't fix them.
A one-day training is Dagen H without the four years.
I know what I'm saying — I run those trainings and I'll keep running them. They do precisely what that September morning did: they focus the attention of an entire organization on a single point. People walk out charged. For two weeks they try things, test things, show each other what they found.
And then the attention runs out. The quarter comes back, the deadlines come back, Excel comes back. The statistics drift back to trend.
Swedish drivers stayed alert for six weeks — and that was a change that could put you in a head-on collision. A team after a one-day AI training stays alert for two.
That's not the fault of the training or the people. It's the arithmetic of attention. A single event cannot hold a new reflex in an organization of hundreds — just as a single morning didn't hold the caution of Swedish drivers.
What holds it is the system standing behind the event. In Sweden that was four years and two tracks. In a company: twelve months and the same two tracks.
That's where AI Growth Initiative comes from.
It isn't a discounted training bundle. It's a twelve-month programme I run inside your organization on two parallel tracks — the concrete one and the habit one. In corporate terms: IT and business.
Because two different people decide about AI at your place, and they rarely sit in the same meeting. One asks whether it complies with GDPR and the AI Act. The other asks whether it shortens the quoting process. Both are right, and both need an answer — otherwise the project stalls and nobody can say why.
Fifteen minutes, no deck and no CRM. You tell me what isn't working at your place, and I tell you whether this programme makes sense there at all.
+48 606 146 588At the start of the month I ask what's going on at your place. Which processes hurt, what changed in the tooling, what people tried and failed at. That becomes the topic of the next session.
Online training, open to anyone who wants to come. No seat limit, no premium and standard groups. 70% of the time is work on participants' real tasks.
A separate session: strategy, priorities for the next quarter, a process audit, tooling decisions. This is where the programme stops being a training and becomes a plan.
Between sessions: what actually changed in AI and what to do with it at your place. Plus a group where you ask a question on Wednesday afternoon instead of waiting four weeks.
Certifications and universities I learned at — and teach at:
Scroll sideways to see the full table
| Ad-hoc trainings | AI Growth Initiative | |
|---|---|---|
| Cost over a year | 4 trainings × PLN 6,000 = PLN 24,000 | PLN 60,000 |
| Time with me | 4 days, disconnected from each other | 96 hours on a monthly rhythm |
| Continuity | Effect fades after 2–3 weeks | A fresh foothold every month |
| Compliance and tooling | Out of scope | A dedicated track from month one |
| Access between sessions | None | Newsletter and closed community |
| ROI measurement | Nothing to measure | Metrics agreed up front, reported quarterly |
| What's left after a year | Handouts and a memory | AI policy, a stack, an internal lead |
Write directly. No form, no details to fill in. I reply within 24 hours.
przemek@pbx.agencyPLN 5,000 net / month
A twelve-month AI rollout run on two tracks at once: IT (compliance, tooling, security) and business (adoption, use cases, ROI). Because one training doesn't change a company — rhythm does.
If it were one training a month, it would be. It's eight hours of work with your organization every month plus everything that happens in between: setting the topic against your real processes, tooling and compliance decisions, the newsletter, answers in the community. Over a year that's PLN 60,000 net for twelve months of continuous work. Four ad-hoc trainings cost PLN 24,000 and what's left after a year is the handouts. If what you want is four training days, buy four training days — it's cheaper and I'd rather tell you straight. The programme makes sense when you want something to remain after the year.
Licences aren't adoption. The picture I see most often: a company pays for several hundred seats, a dozen or so people log in, and procurement asks why we're renewing. The tool is usually the smallest problem — what's missing is rules (what may go into a model), process (where exactly this shortens work) and someone to ask on a Wednesday afternoon. If you already have the licences, so much the better: we spend nothing on new tools and pull value out of the ones you're already paying for.
Three people, and less time than you'd expect. Someone from HR or L&D as the host of the programme — they gather needs and keep the dates. One person from IT for the compliance and tooling conversations. And a board sponsor who makes decisions once a quarter. Realistically: two to three hours a month on your side outside the sessions themselves.
Then I bring the topic. Before each month I ask what's going on at your place, and that becomes the session. And if genuinely nothing is in the air, I have modules that work in any organization: working with documents, no-code automation, AI in meetings and reporting, data hygiene and security. There has never been a month with nothing to talk about.
The training component usually can, but a monthly subscription can be awkward to fit the paperwork, and I won't promise you something I don't control. I don't file the applications myself — that's a separate specialism. I refer you to a partner who has handled BUR, KFS and EU funds for years, and on the first call we check what can be put under a subsidy and what is simply cleaner to invoice.
Then we stop. There's no minimum term and no cancellation penalty — it runs month to month. I know how that sounds for something I sell as a twelve-month programme, but I have a simple problem with the alternative: if I had to hold you with a clause, the programme isn't delivering. I'd rather you had a reason to stay every month.
Sweden did not change which side of the road it drove on, on 3 September 1967.
It changed it over the four years before — sign by sign, bus by bus, campaign by campaign. On the third of September it merely crossed over. That was the last and shortest part of the job.
Companies rolling out AI start on the third of September. They buy the licences, run the training, announce the transformation — and wonder why a month later everyone is driving the old way.
I suggest starting with the four years. Compressed into twelve months.
Tell me where you are. I'll reply within 24 hours and say plainly whether I see a fit — including when I don't.