The Windstorm

A weekly dispatch from the eye of the storm

Week of August 3, 2026

Welcome to the Eye of the Windstorm.

Anthropic shipped Claude Opus 5 on July 24, and the headline everyone repeated was the price: frontier intelligence at half the cost. That is the easy part of the story, and it is worth saying plainly - the easy part keeps getting easier. The switch on the wall got cheaper to flip again. What did not change, what never changes, is whether anyone in your building knows what to do once the lights come on. The tool was never the hard part. You are.

1. Opus 5, and the falling price of genius

On July 24, Anthropic released Claude Opus 5 at the same price as the model before it - five dollars per million tokens in, twenty-five out - while more than doubling its predecessor's score on frontier coding work and landing near the top of the pack. Translate that out of benchmark-speak: the smartest advisor you can rent got noticeably smarter, and your bill stayed flat. If you have been waiting for the tools to be good enough, that excuse quietly expired this week. They are good enough. They have been for a while.

2. The 88 percent nobody puts on a slide

Here is the number that should be on every executive's whiteboard. IDC's research puts it at roughly 88 percent of AI agent pilots never making it into production. And the reasons given are almost never the model — they are data readiness, governance, evaluation, and plain integration. The agents that do cross the line? They average a 171 percent return. Read that twice. The technology is almost never what's broken. The wiring around it is.

3. Everyone bought the electricity

Ninety-seven percent of executives say their company rolled out AI agents in the past year. Roughly half of employees are touching them. But only about a third of companies have actually begun scaling any of it across the organization, and Gartner expects 40 percent of enterprise applications to embed task-specific agents by year's end, up from under 5 percent a year ago. So the current is running through the walls. The meters are spinning. Almost nobody has taught the whole building how to flip the switch.

The Eye

Step back from the noise and look at what the week actually told us. The price of raw intelligence fell by half, and 88 percent of companies still can't get it to do useful work. Those two facts sit right next to each other, and together they close the case: capability was never the bottleneck. The bottleneck is judgment. It is knowing which question to ask, which of your suddenly-brilliant advisors to trust on a given morning, and - hardest of all - what to keep and what to throw away. That last one is the great winnowing skill of this decade, and I'll say it again: nobody is a virtuoso yet, not the labs, not the consultants, not me.

Think of yourself as a child monarch who just inherited a table of counselors, each one knowing a thousand times what you do about their corner of the world. You will never out-know them, and you don't have to. Your whole job is to ask the right question and choose well among the answers. The companies winning right now aren't the ones with the best model - everyone rents the same model. They're the ones who taught their people to conduct instead of to compute. That is human work. It always was.

Do this with your agent this week

Pick one decision you make on repeat - a hire, a vendor, a pricing call - and don't ask your agent to make it. Turn it around. Type: "Before I decide, ask me the five questions a world-class analyst would ask about this, one at a time, and wait for my answer before the next." Then actually answer them. You'll feel the muscle you've been skipping. The superpower was never the answer. It was the question.

From the eye of the storm,
Grant