A week of practice on your own reports and summaries. You build a check in which a second agent and a few simple cross-checks look for the mistakes, and you only review what they flag.
The English edition opens once there’s a group. It’s for people who already use AI at work.
According to a 2025 market study, packaging prices rose 18%.Made up: there is no such study in the attached files
We compared 14 suppliers on price and lead time.Count: the table has 12 rows
All delivery times are within the normal range.Lost: the email from supplier Nord mentions a six-week delay
We also assessed the risks of entering Asian markets.Brief ignored: you asked for a price comparison only
This is how managers describe working with AI, in our graduates’ conversations and in a survey of companies in Estonia. If one of these sounds like you, this course is for you.
I check everything it does by hand. In the end it isn’t faster than doing it myself.
Some of the research it cites and some of the numbers are simply made up.
It slips its own content in where you least expect it, so I reread the whole thing every time.
My manager found the mistake in my report before I did.
I asked it to check its own work. It told me, very confidently, that everything was correct.
Getting the numbers right in a report from several sources takes round after round.
Nobody expects a new hire to make no mistakes. A manager agrees what gets cross-checked, who takes a second look and when an issue comes back to them. AI works the same way, except the second pair of eyes is a different model, and it takes minutes.
Recount the numbers, find the quote in the source, check the result against the list of requirements. No second model, and hardly any of your time.
A different model with one job: find what isn’t in the source files, what breaks the brief and what got lost along the way.
You look only at what the first two steps flagged, and decide: fix it, let it go, or dig in yourself.
Why a different model does the checking: in measurements, a model is right about half the time but sure of itself far more often. Asking it to recheck its own work does little.
You do every assignment on AI output you get regularly at work: a report, a summary, research, a calculation. By the end of the week, its check is ready.
Find the first mistake in a recent piece of AI output of yours, and time how long it took.
A made-up fact, a broken requirement, a lost point, a wrong calculation: mark which ones threaten your report.
Write a checklist of 3–5 cross-checks that don’t need a second model.
Write the brief for the checker and compare what it found with what you found by hand.
Decide what in your report can go out after the check, and what you always look at yourself.
Put it all into one set of instructions and run it on a fresh piece of work.
Two browser tabs, two different models, everything by copy and paste. You can start the same day.
An agent in the terminal reads the source files, runs the checker and recounts the numbers by itself. We show it as the next step, once the basic level works.
I expected the course to sell me on AI in every shape and form. In reality, half of it is about the mistakes AI makes – and the mistakes people make reading its results.
Mikhail Seleznev, Project Manager, PMP, Bereke Bank, on the Russian-language course “AI for managers”
Free material explains why AI makes mistakes. Here you build a check for your own work and test it on your own output, and your assignments are reviewed. You leave with a routine you run every week, not a list of tips.
No. The basic level is two browser tabs with two different models. The agent in the terminal is an optional next step.
Any two different ones, for example Claude and ChatGPT. What matters is that the model doing the checking isn’t the one that did the work.
A second agent costs time and tokens, so the course has a lesson on where checking pays off and where it costs more than doing it by hand.
Once there’s a group. People on the list hear first, with the date and the terms.
The English edition opens once there’s a group
Course price: 6 000 ₽35 000 ₸€65$75
While you wait: our research on how AI models handle a manager’s work is already in English. See the model rankings
We’ll write when enrolment opens. Meanwhile, see how models hold their position under pressure, from our research.
Open the researchWe use cookies: technical ones keep the site working, analytics ones show us what works on it. You can turn analytics off. Learn more
