The 2,000 AI Use Cases That Told Me Nothing I Needed To Know

Someone on LinkedIn offered me 2,000 enterprise AI use cases. Of course I clicked.

What I got was a list of fifteen links — consultancies, hyperscalers, software vendors, a couple of PDFs. The promise was scale: two thousand things you could do with AI inside a business. I spent some hours reading them. Not all of them cover to cover — nobody has that kind of time and nobody should — but enough of each to understand what was actually on offer.

Here is what I found. Two of the fifteen were doing real work. Google and BCG had built a bottom-up market sizing across roughly 330 task categories, landing on a claimed trillion-dollar services opportunity. OpenAI had written a practical guide built around six types of use case and a framework for prioritising them. Useful, in different ways, to different readers.

The other thirteen were product catalogues wearing the language of market research. Every Salesforce use case involved Salesforce. Every SAP one involved SAP. The Google Cloud list of “1,001 real-world use cases” turned out to be a running tally of customer name-drops, one or two sentences each, started at 101 in April 2024 and topped up ten times since. The McKinsey piece being recycled into the 2026 collection was written in February 2024. Two years old, undated in the offer. That is the inventory.

The thing that struck me

Across fifteen sources, across every major consultancy and every major enterprise software vendor, not one of them engaged with the question of what happens when AI arrives inside an organisation without being chosen.

Every document in that collection is organised around the same assumption: that AI is something a company decides to adopt, sold — by the authors of these very documents — on its benefits. You identify a use case. You build a business case. You procure, you deploy, you measure the return. It is the traditional model of enterprise technology adoption, and it is how boards and executives are being taught to think about AI right now.

It is not how AI is actually arriving.

Inside the organisations I work with, the AI that matters is not the AI anyone chose. It is the AI that turned up. It arrived inside a platform update. It was enabled by default in software already licensed. It was built inside the building, in plain English, by someone who does not report to IT. It was delivered by a system integrator as part of something else. None of that shows up in a use-case inventory. Because a use-case inventory is a description of what you could buy. It is not a description of what you already have.

The asymmetry

That is the asymmetry. On one side of it, an entire industry is organised around selling AI as a decision. On the other side, inside the walls of the organisations being sold to, AI is behaving like weather. It arrives. It settles. It changes the climate of the business before anyone in the boardroom has used the word.

Boards are being handed maps. The maps describe the AI they could acquire. They do not describe the AI already inside the building. This is not a claim that the vendors are hiding anything. The documents do what they say they do. They describe what is for sale. The gap is not in what they contain. The gap is in what they are not describing — which is the part of the AI estate that did not come through the front door.

The question to take into the room

So here is the question I would take into the next executive conversation where someone puts a use-case inventory on the table. Which of these use cases is already running inside our organisation by default, and who would know? What do we have? Do you know? How do we find out?

Because until you can answer those, the inventory is not the thing that matters.