The Law

The article arrived the way these things do: a newsletter, a Thursday morning, a title scanned between two meetings. Then the title registered.

Beyond the copilot.

McKinsey had surveyed 334 product and engineering leaders about AI. The finding that stops you: three in ten organizations saw productivity fall after adopting AI tools. Not stagnate. Fall. Meanwhile a small group pulled far ahead, and what set them apart was not which tools they bought. It was that they redesigned the work itself. Their conclusion, in one line: "technology is not the differentiator."

We have been saying a version of that sentence since day one. The gap is not technical. The gap is architectural.

An opinion needs defending. A law just needs time.

When we said it, it was a position. A founder's conviction, the kind you're supposed to have. You could nod politely and move on.

That's the difference between an opinion and a law. An opinion needs defending. A law just needs time.

What McKinsey documented is the law at work: give a copilot to an unchanged process, and the process stays unchanged. The tool suggests. You still drive. You still read, decide, forward, wait. The screens get smarter; the outcomes don't. Some organizations even go backwards, because now there's a tool to manage on top of everything else.

The organizations that pulled ahead did something structurally different. They stopped asking what can this tool do for our process and started asking what should the process be, now that agents exist. Agents run the flow continuously, around the clock. People appear where judgment is needed, and nowhere else.

That is not adoption. That is architecture.

The same law, a different domain

McKinsey writes about how software gets built. We build how claims get handled. Different domain. Same law.

A claims process with a copilot is still a claims process: a human reads the report, decides, forwards, waits for the garage, chases the file. The copilot drafts the email faster. The waiting is untouched.

A claims process rebuilt around AI looks like nothing that came before it. A driver hits a pole at 8:47 and sends a photo. Wilma receives it, confirms in his language, validates the estimate, books the repair, keeps everyone informed until the car comes back. The fleet manager hears about it in a pattern report, if there's a pattern worth hearing about.

The expert doesn't disappear. The expert appears exactly where a human eye matters, and nowhere else. That is the blended system McKinsey describes: people and agents, each with defined responsibilities, organized around a redesigned flow of work.

We didn't read their article as a roadmap. We read it as a map of where we already live.

The trap has a name now

There's a number we've used before: 73 percent of insurers run an AI initiative, 7 percent scale it. We called it pilot purgatory. McKinsey's survey draws the same shape from a different angle: tool adopters rarely break through; the ones that redesign roles and workflows do.

Two datasets. One pattern. Adding AI doesn't work. Rebuilding around it does.

The comfortable reading of that pattern is: eventually everyone will get there. The uncomfortable reading is the true one: the longer a process is merely assisted, the more the assisted version calcifies. Every copilot bought is another year the old architecture survives.

We didn't read it as news

Here's the honest part. It's tempting, when the world's most quoted consultancy publishes your thesis, to take a bow. We'll resist. We didn't predict anything. We just made a choice at the foundation, eighteen months before the survey data existed to justify it.

Some choices you make because the evidence tells you to.
This one we made because we could see what it becomes.

The evidence caught up. It usually does. That's the thing about laws.

---