Want to see my spreadsheet?

I sometimes make a joke at the beginning of my workshops.

“Show of hands, who loves Excel?”

There is always at least one hand. Usually there are two. Random people who love Excel enough to overcome the hesitation of participation and raise their hand together with me.

I understand them.

There is a particular kind of satisfaction in building a spreadsheet that works exactly the way your mind works. You Excel lovers know the feeling. You open a blank workbook, start naming tabs, create a few formulas, adjust the formatting, and it emerges. Your fingers know the shortcuts, time flies by.

It evolves over time. First it is a calculator. Perhaps a memory aid. A scratch pad. A model of the work as you understand it. You’ve learned not to format it too soon, as you’ll just end up doing it again.

People get attached because the spreadsheet becomes an extension of how they work and think.

In the workshops I then ask the second question.

“Now, who loves working on Excel with someone else?”

That one usually gets a laugh. Or at least a snicker. Everyone knows. Excel is wonderful until someone else has to use it with you.

You build the sheet the way you think. You name things in a way that makes sense to you. You create tabs according to your own logic. You use formulas you understand, or formulas you looked up and hope you got correct. You color-code just enough to make yourself feel organized.

And then someone else opens it.

What felt obvious to you is not obvious to them.

They do not know which tab matters. They do not know which cells are inputs. They do not know which formulas are fragile. They do not know whether the number they are looking at is current, historical, experimental, or something you copied over during a meeting and forgot to clean up. We are proud of our spreadsheets, even though we understand that no one else will fully recognize the brilliance, the subtlety, the value we created.

Using LLMs is similar. It is part of why LLMs feel so magical at times (when they work).

They meet us where we are.

You don’t have to remember an obscure function. You do not have to know the exact syntax. You do not have to structure your question perfectly before you begin.

You can describe what you want in your own language, revise as you go, ask follow-up questions, change direction, and build toward the answer in a way that feels natural. The tool adapts to you.

It can seem miraculous. Yet the collaboration challenge is there, just as with Excel, hovering in the background. And the results, the outcomes, the insight we’re so excited about, cannot be fully trusted. There is a layer, just beneath the surface, that needs to be double-checked. Confirmed.

A lot like Excel formulas.

Organizations do not just need individuals to get better answers. We need someone to cross-check. We need collaboration. Shared definitions and shared processes. Shared data and accountability. Shared confidence that the output is useful, safe, accurate enough, and connected to the way the organization actually works.

Alignment, in a word.

Trust, as well, if I get two words.

And here we are, back in collaboration mode. Tough in Excel and LLMs. The individual tool is not enough. AI products have to do more than just wrap a chat.

Suddenly we’re all product managers.

Product managers ask tough questions. Who owns the prompt? Who owns the data? Who checks the answer? How do we know if it was checked? Who updates the process when the model changes? Who notices when two teams are using the same tool in completely different ways and calling the result the same thing?

Having access to AI does not remove the need for process thinking. It brings it back in spades, but this time in a new form, with a twist. Because we cannot trust the output.

That does not mean the output is useless. But an AI model is not a colleague we have worked with for years. It lacks the consistency and predictability we learn to expect from human colleagues. It may produce something useful, plausible, incomplete, or wrong, all with the same confidence, and we have no easy way to tell which is which.

That is the product design challenge now. Building systems that still work when people have to use them together. Accommodating that you cannot trust the AI model the same way you can trust a colleague.

I still love a good spreadsheet.

I love the moment when the tabs make sense, the formulas work, and the page reflects the shape of the problem in my own head.

But I rarely share them.

I know how people feel about my spreadsheets by now.