Inside Companies Adopting AI. The Technology is not the Problem.

When a platform detects slop, it flags it. When your buyer detects it in your content, your proposal, your deck — they just quietly trust you less.

It was my birthday this weekend. And beside enjoying cakes, friends and family, I was thinking about a strange fact: in the last months of consulting I’ve seen the inside of more companies than in my previous decade as an operator, and almost none of their AI problems were about AI.

In January I left operating — one team, one P&L, one roadmap — to consult full-time.

Since then I’ve worked with distributors, manufacturers, SaaS platforms and retailers. All of them implementing AI somewhere in their commerce operations. All of them convinced the hard part was choosing the right tools.

A company would bring me in with a shortlist of AI vendors already drawn up, and within two weeks it was obvious the shortlist was beside the point. The product data didn’t connect to the search experience. The search data never reached forecasting. And somewhere in the building, a team was shipping AI-generated work that nobody, including its author, had actually read.

The number I kept living inside

MIT put a figure on it this year: 95% of enterprise AI pilots show no measurable P&L impact. Thirty to forty billion dollars spent, ninety-five percent invisible.

BCG traced most transformation failures to organizational culture rather than technology.

Neither finding surprised me. I’d been watching it happen in conference rooms all spring.

Workslop

Researchers at Stanford and BetterUp gave the phenomenon a name I now use with clients: workslop. AI-generated work that looks polished but doesn’t advance anything.

Forty percent of workers received some in the past month. Each piece costs its recipient nearly two hours of cleanup.

The output looks like progress. The work never happened.

Inside a commerce team, workslop has a particular flavor. Product descriptions generated in bulk and shipped unread, which customer service then spends weeks compensating for. Strategy decks produced in twenty minutes and approved by leadership, unexecutable because they were never connected to the data the company actually has.

What makes it dangerous is that everyone involved believes they’re moving faster. The velocity is real. The progress isn’t.

What the winners had in common

It wasn’t budget.

The companies getting results were the ones that understood their commerce operation as a single connected system — product data feeding search, search feeding conversion, conversion feeding forecasting, human judgment sitting on top of all of it.

Before automating anything, they asked which parts of the work still required someone to think.

Everyone else automated the thinking and kept the busywork.
The same Stanford and BetterUp team found this split runs through individuals, too.

Across 12,000 workers, they identified “pilots,” who use AI to do more thinking, and “passengers,” who use it to skip the work. Pilots turned out to be 3.6 times more productive. Same tools.

David Brooks made the larger point in The Atlantic this summer: when intelligence is plentiful, volition is valuable.

Every company I visited this spring had access to the same models. The difference was never the software. It was whether the team still wanted to think.

The button

Last week LinkedIn added a button to its feed: “Seems like AI slop.” A detection firm had just found that over 40% of long posts on the platform were fully machine-written, so LinkedIn is now paying to figure out which of its users still think for themselves.

That’s the part worth sitting with. A social platform can build a slop detector. Your customers can’t.

When a buyer receives your AI-generated product content, your AI-written proposal, your auto-assembled deck, there’s no button — there’s just a quiet decision to trust you a little less. The platforms are only making visible what’s been happening inside companies all year.

Most companies I visited this spring could have used that button in their Monday meetings.

Six months in

The first thing I check with a new client is no longer their tech stack.

It’s how work moves between people, and where the thinking still happens.

That’s where the AI results are hiding.