AI Adoption Does Not Start With a Tool. It Starts With a Way of Working

A practical view of AI adoption in companies: why choosing a tool is not enough, and what needs to change in how work is briefed, tested, and decided.

When AI comes up inside a company, the conversation often starts with tools. Which model should we choose? Should we buy team licences? Should we allow ChatGPT, Claude, Copilot, or another platform? This discussion is necessary, but if the company stops there, it will probably just add another app to the working day.

The real shift happens elsewhere. In how people brief work, how they prepare first versions, how they test ideas, and how they decide what should move forward.

Denisa Hrubešová from Forendors showed this clearly at Women in AI Prague. She did not talk about AI as a nice add-on to work. She described a change in her own role. As a CEO, she no longer only waits for someone else to prepare the technical execution. She can quickly create a first draft, review options, discuss risks with the team, and only then decide what makes sense to move forward.

That is practical AI adoption. Not a statement that the company uses AI. More of a daily change in the rhythm of work.

Companies often ask the wrong first question

The question “Which tool should we implement?” is comfortable. You can build a comparison table, choose a vendor, and mark the decision as done.

But then the team often does not know what to do with the tool. People try a few prompts; some work, some do not. Someone is excited, someone is sceptical, someone is afraid of making a mistake. After a few weeks, AI stays with the most active individuals, and the rest of the company continues as before.

A more useful question is: where are we currently waiting too long for a first version?

It might be a product idea, an internal report, a campaign concept, a training proposal, a document structure, an analysis of feedback, or the first draft of customer communication. This is often where AI has the most value. Not because it decides for the team, but because it shortens the path to something concrete that people can discuss.

The first version no longer has to wait for another person

In many companies, the delay starts before the actual work begins. Someone has an idea but cannot name it clearly. Another person needs a brief but receives only a vague feeling. Then the first proposal is produced, feedback comes back, the brief is rewritten, and only after several rounds does it become clear that people imagined different things from the beginning.

AI can help here as an intermediate working step. A person can enter a raw thought, a voice note, a screenshot, a few brand rules, or a description of the problem. The tool can turn that into options, an outline, a prototype, or a proposed solution. The team then discusses something concrete rather than an abstract idea.

The important part is that the first version is not the final truth. It is material for discussion.

A new way of working has four steps

A useful AI workflow has a simple logic.

First comes the input. It does not have to be perfect. It can be a sentence such as: “I do not know exactly what this should look like, but this is the problem.” This is where many people get stuck unnecessarily because they think they need to write the perfect prompt. They do not. A rough description of the problem is often enough to start.

Then comes processing. It might be Claude Code, Codex, a custom agent, a custom GPT, or another tool. The name of the tool matters less than whether it has enough context: who the output is for, what the constraints are, what tone it should use, what it should ignore, and where it should name risks.

The third step is the output. Not a final document, but a first version: a prototype, solution options, a draft brief, an article structure, data hypotheses, or a list of questions for the team.

The fourth step is review. This is where the difference between reasonable AI adoption and chaotic tool use becomes visible. A human needs to validate UX, risks, priorities, technical feasibility, security, and the impact on the customer or the team.

The human stays with the decision

The biggest mistake in AI adoption is pretending that the tool takes over responsibility. It does not. It can speed up preparation, summarise data, suggest options, highlight patterns, and help with a first version. But the decision remains with the human.

That is also a practical change in the role. Less time goes into manually assembling the first version. More attention shifts to interpretation, selection, review, and accountability.

For leaders, the conclusion is simple: AI adoption does not need to start with a large programme. It is better to pick one specific working moment where there is currently waiting, ambiguity, or unnecessary rewriting. That is where the new working mode can be tested fastest.

A practical test for one idea

Take one idea that has been sitting in your head or in Slack for a week. Dictate or write:

I do not know exactly what this should look like. This is the problem. Suggest three possible solutions. For each one, add the main risk, what a human should check, and what decision we need to make before the next step.

Then do not look only at the quality of the answer. Watch what happens to the work. Did the conversation move forward? Did the team understand the problem faster? Was it easier to say what you do not want? Did a better question emerge?

That is where practical AI adoption begins.