Value gap, not model gap

Gartner expects more than 40 percent of agentic AI projects to be cancelled by the end of 2027.

Not because the models were bad. They rarely were.

The model gap is a technical problem. It shrinks month by month. The value gap is a human problem. It stays.

Last week, a software company I advised, owned by its own customers. The mandate arrived as a wish:

"We need to get our costs down."

The same situation, as a problem:

"Every customer fix waits up to twelve months for the next annual release. We'd know it's solved when a fix reaches customers in the same quarter it was reported."

The second gate is the number that decides whether it continues — the line that was missing:

"If the first quarterly release hasn't reached a customer by a date we set now, we stop and go back to the cost case."

That third line isn't a target. It's permission to stop — given in advance, by everyone, while nobody's reputation is attached to it yet.

In that company, the line held. The annual cycle went away, and fixes reached customers quarterly. Which is why I know the number was never the hard part.

One move: before the pilot starts, write down the number that would stop it, and the date you'll look.

But be careful which number. Most AI mandates are written for fewer people and more output, so the number gets taken from the old place — the process and its unit cost. And no unit cost ever showed why the work landed.

That is where the machine's half shows up. The half that decides whether any of it works does not.

KPMG measured 77 percent versus 20 percent positive business outcomes, depending on how confident a company was in its own people. Not in its models.

Intent isn't a workshop. It's two sentences, written down at two moments: before it starts, and before it's allowed to continue.

Next Tuesday: written down is not the same as heard.

💬 Has anyone here stopped an AI pilot on purpose — and what made that possible?