Uhlík vs Křemík · Czech podcast about artificial intelligence

Petr Kohout: Public offices need AI for a hundred people, not four enthusiasts

Episode 15 · 5 August 2026 · 57:04 · Guest: Petr Kohout

Listen on: YouTube · Spotify · Apple Podcasts

Petr Kohout, director of the legal department at the Czech Ministry of Health, is one of the most visible advocates of artificial intelligence in public administration. In episode 15 of the Uhlík vs Křemík podcast with David Salomon and Adam Hanka, he explains why he cares less about the newest models than about whether ordinary civil servants use AI. He also talks about the fear of admitting to working with AI, about failing pilots and about the role of lawyers.

Mass use instead of novelties

Kohout joined the ministry on 1 July after years at the Regional Authority of the Central Bohemian Region. He says he has stopped caring about cutting-edge novelties and is returning to the basics. What matters more to him is implementation at mass scale: "For me it is better when a hundred users use Copilot or anything else for 20 minutes a day than when I see four people working with it nonstop."

He also points to what he calls digital feudalism. Czech companies and public offices run on American solutions, from Office and Google to Amazon's cloud, and if it suited someone in America, they could be cut off. In his view this is not tied to artificial intelligence alone. Organisations should therefore build critical processes so that, in the worst case, the model underneath can be swapped. Adam Hanka added that he is trying the French Mistral in his own pipelines and that sources need to be diversified.

Tokens do not measure value

The debate turned to how to measure the value of AI. Some companies track how many tokens their employees consume. Adam Hanka sees this as a metric that mainly reveals people who do not use AI and need support. Kohout adds that the basic question of what an organisation actually wants to use AI for is often missing. He recommends writing down on an A4 sheet what you want to achieve and admitting that a simple task only needs an ordinary robot.

Who should drive adoption in a public office? According to Kohout it has to come from the top and from below: leadership that knows what it wants, and ambassadors who show colleagues over coffee that a tool can do the tedious work. Digital debt is a brake. A lot of digitalisation, he says, simply converts a paper process one to one into digital form, and the result then takes longer than on paper.

The fear of admitting AI helps

According to Kohout, there is still a fear in public administration of admitting to using AI. It might be reports for management that would otherwise take three hours to write and are done in five minutes with the tool. People fear the question of what the office needs them for then. Kohout answers with the example of lawyers: "He is paid for the mental capacity, for the performance." Specialised tools do research and interpretations, but a lawyer has to recognise hallucinations and case law that looks good and still does not fit. With better models, he says, this keeps getting harder.

A new part of the job, he says, is defending your opinion against a ChatGPT answer a colleague prepared in advance. It tends to be the right answer to an incompletely asked question. He himself uses AI mainly for his calendar, tasks, summarising emails in Copilot and dictating tasks into Gemini on the way back from meetings. He does not put personal data into the tools, and he cannot paste a court letter with people's data into AI.

Why pilots fail and why lawyers should lead

One reason pilot projects fail, according to Kohout, is the amount of work on the client's side. A supplier promises a solution in 30 hours, but then needs data and explanations of processes, so the office ends up working perhaps five times as many hours internally. Adam Hanka knows this from practice: in his own words, he spends the first month at a client mostly handing out homework.

At the Ministry of Health, Kohout values responsible implementation and support for innovative solutions and startups. Healthcare needs AI because there will not be more people working in it. And lawyers, he believes, belong on the front line, because regulation has to be prepared by people who understand the technology: "artificial intelligence is not a technological revolution, for me artificial intelligence is a sociocultural phenomenon."

The full conversation with Petr Kohout is available in the Uhlík vs Křemík podcast.

The episode is in Czech. Czech page with chapters, quotes and the full transcript