The Windstorm

A weekly dispatch from the eye of the storm

Week of July 27, 2026

Welcome to the Eye of the Windstorm.

This week the numbers finally said the quiet part out loud: 97% of companies deployed AI agents in the past year, and fewer than 10% of their employees are equipped to actually use them. Only 3.8% of workers can write the instructions to build a simple automation. Read those two facts side by side and you have the whole story of this decade. The tools arrived. The people did not arrive with them. That gap is not a technical problem you can buy your way out of — and pretending otherwise is the most expensive mistake being made in business right now.

1. Everyone deployed. Almost nobody is equipped.

The new adoption data is stark. Ninety-seven percent of executives say their company put agents into production this year; 52% of employees have touched one. But 73% of the workforce are still what researchers call "Experimenters" — people who use AI for the occasional one-off and then close the tab. Fewer than 10% of workers can even define an agent in their own words. And the punchline that should keep every CEO up at night: fewer than 10% of companies have scaled these tools into anything you could call tangible value. The buying spree was easy. The wielding is where the whole thing is stuck.

2. Claude Opus 5 made the tools cheaper — again.

On July 24, Anthropic shipped Claude Opus 5: near-frontier reasoning at half the price of its top model, a million-token context window, built specifically for long-running, multi-step agent work. Box tested it and found a 17% jump on due-diligence workflows and 11% on data analysis over the prior model. Here is what I want you to notice: the capability got better and cheaper in the same breath, for the fourth or fifth time this year. The tools are not the bottleneck. They have never been the bottleneck. Every month the electricity gets cheaper, and every month the building stays un-wired.

3. The great cancellation is coming.

Gartner now forecasts that more than 40% of agentic AI projects will be scrapped by 2027 — killed by unclear ROI, runaway costs, and controls nobody built in at the start. Today only 41% of agent rollouts cross into positive return within a year, and 19% never pay back at all. Underneath sits a 60% governance gap: agents are in production, and the judgment about how to run them is nowhere to be found. This is the winnowing arriving on schedule. The companies that survive it will not be the ones who deployed the most. They will be the ones who chose the fewest, right bets and knew why.

The Eye

When we electrified the factories, the payoff took forty years to show up in the numbers — not because the electricity was slow, but because the buildings were built around steam. You had to tear out the driveshafts, rewire the floor, and teach every worker on it to flip a switch instead of throw a belt. Buying the electricity was never the hard part. This time forty years became four, and the same law holds: the model is the easy purchase, and the human rewiring is the whole job. That is why 97% can deploy and fewer than 10% can profit.

So stop asking your people to be experts. You are a child monarch at a table of advisors who each know a thousand times what you do — and the throne was never won by knowing more than them. It was won by asking the sharpest question in the room and listening to which advisor flinched. The superpower of this decade is not the answer; it is the question, and then the winnowing of a hundred plausible answers down to the one that is true. Nobody is a virtuoso at it yet. That is the good news. The seat is still open.

Do this with your agent this week

Pick one task you did by hand this week and did not enjoy. Before you touch a single setting, write your agent a plain-English brief: what "done" looks like, what it must never do, and one example of a good result. Then hand it over and grade what comes back like an editor, not a user — mark what is wrong, hand it back, and watch it improve. You are not learning a tool. You are practicing the one skill the 3.8% already have: telling a very capable stranger exactly what you want.

From the eye of the storm,
Grant