Excel Didn't Make Every Company Better. Neither Will AI.

A while ago I ran a small experiment: I asked an AI to build me a slide deck explaining how a boxer engine works.

It simply produced one. Structure, captions, layout — without being fed source material, without picking a template, without learning another piece of software.

A year ago, that took a dedicated course.

How much of last year's curriculum is left?

Think back to what AI courses taught a year ago: how to write prompts, how to generate songs with a music tool, how to produce decks with a slide tool. The classrooms were full, and people studied hard.

This is not a complaint about those courses. The iteration speed of the tools simply outran the half-life of the curriculum. What needed dedicated training then is mostly a built-in ability of general-purpose AI now — and even the slide deck as a format is loosening its grip. In many settings, a single scrollable page explains things better than a stack of slides.

Two months of client work this summer made the shift concrete: it is no longer about how well you prompt. If you can describe what you want clearly, the AI can build it.

What required a dedicated course a year ago is a built-in feature of general-purpose AI today.

The barrier moved. It didn't disappear.

We have seen this script before.

When Excel first spread through offices, everyone could manage a simple table, but the people who could write deep formulas or working macros were rare — one or two in a company, if any. There was a visible wall between those who could and those who couldn't.

AI tore that wall down. Anyone can now step into water that used to require a specialist — writing software, building systems, running analysis. If you can say it, you can have it.

But the years when Excel reached every office did not make every Excel-using company better. Some used it to bring real order to their management. Others just moved the chaos from paper into cells. Same tool, two different outcomes.

The barrier to using the tools came down. The barrier to judgment stayed where it was — and with more options than ever, it got higher.

The dividing line we see on-site

In two months of recent projects, the tool was never where things got stuck. Things got stuck earlier: in stating what was actually wanted.

Where does the process break? Which step drains the most people? Once it changes, who takes over? Owners who can answer those questions feel the difference the moment a tool is connected. Owners who can't will find that a stronger tool only runs the existing chaos faster.

This is the same ordering I described in Before you add AI, find the process that is really stuck. The stronger the tools get, the more that ordering matters — because the tools are now strong enough to build the wrong thing beautifully. A mis-described requirement gets executed quickly, and very well.

Which is why Winnow doesn't sell tools

Winnow does AI process improvement, and I deliberately set it up as a consultancy, not a technology vendor. This is the reason.

Tools keep being replaced. The tool chosen this year may be outclassed next year; the operating skills a team learns now may be irrelevant the year after. If what I sold were tool operation, the shelf life of this business would be one iteration cycle.

What a consultant sells is the part that survives the tool change: understand how the business works first, then judge where technology should step in. From pen and paper to Excel to AI, the tools have turned over three times. The judgment has been the same job all along.

The company's English name is Winnow — the act of tossing grain into the wind so the chaff scatters and the kernels fall back. These waves of AI tools are the wind: they arrive in gusts and leave in gusts. What stays in your hands was never the tool. It is what the wind has sifted clean — your understanding of your own business.

When the next wave of tools arrives, the question worth asking first is not "what can this tool do?" It is "can I say, clearly, what I want it to do?"