A demo puts on one screen the result of choices and corrections that are not always visible. Next to it, your first attempt looks small. To really use AI at work, start with a need you know well and can check.
Using AI at work goes beyond asking for text. It can mean turning material into a usable deliverable — a visualization, a comparison, a small interface — and getting it to the meeting, the document or the person who will use it. The learning starts when you open the result, test it and say what needs to change.
Pick a first deliverable you can evaluate
If you know a recurring meeting well, you can organize the materials that come before it. If you follow a process, you can explore a screen that makes the information clearer. Familiarity with the subject helps you spot errors and gaps.
Picture a manager who wants to compare support requests from different teams. The first version can put criteria and data in a table, keeping the empty fields visible. There is no need to start with a full approval system.
Use approved materials and a limited sample. Say what the deliverable needs to make possible and what you will use to check its quality.
Treat review as part of using it
Open the result and walk through a situation you know. When something fails, describe the difference between what happened and what you expected. When it works, note which information made the request clearer.
This practice builds your own sense of what you can guide and where you need help. It may also show that the need was simpler, or different, than it seemed.
Get the answer into the work
You get a useful analysis and close the conversation. Days later, the idea is still sitting in the chat history, never having reached the meeting or the person who needed it. Between an answer and its use there is a handoff worth planning.
Before you ask, describe where the deliverable is going: who will use it, on what occasion and to do what. If the team asked for ideas to improve customer service, the next step might be to turn the list into an agenda — which problem each idea solves, what information supports it and what the team needs to confirm.
Before you end the session, note who will take the next action. Not every suggestion becomes a project: some help rule out a path, and recording the conclusion is enough.
Come back to the deliverable after first use
Did it help the conversation, or did it need rebuilding? What was missing for the person who received it? These observations improve the next request. An answer finds its place in the work when someone can use it for a clear purpose.
For a leader or a founder, taking part in this first cycle improves conversations with the team. It brings concrete examples of possibilities, limits and review criteria, without having to take on the technical implementation.
AI Thinking, Dani Hayoshi's method, organizes this path into three moves: question what needs to be understood, redesign what makes sense and rebuild the choice in use. For other ways of working with AI:
AI working on your files: how to oversee the deliverable
Voice and AI: how to test new ways of working
What a first AI-built tool needs to prove
Dani Hayoshi is the author of AI Thinking. Her work connects artificial intelligence, education and entrepreneurship. About Dani.