Newsletter – Edition 4
A quick update on the book. It is in design, and the cover is coming together. I am going back and forth with the publisher’s design team, so nothing is final yet. When the cover is ready, you will see it here first.
Which brings me to a favor.
In Edition 2 I told you about the subtitle I almost kept, “What Wins When Everyone Can Go Fast,” before I changed it to “The Competitive Advantage AI Can’t Buy.” The old line, which is part of the cover image for this newsletter, did not make the book’s cover. It has now found another home, hopefully.
I pitched a book reading session for SXSW 2027, and the talk carries that original subtitle. It is the argument of the book for a room of leaders: where your real AI advantage now lives, why the price of autonomy is coordination, and a simple way to sort what to automate, what to augment, and what to keep human.
SXSW picks part of its program through a public vote called PanelPicker. Community votes are one of the things the organizers weigh. If you have two minutes, a vote would mean a lot to me.
Vote here. Voting is open now and closes August 23. You may need a free SXSW account to cast it.
The week in ideas
Four posts from the past week.
Why Ford Rehired. Ford added more than 350 experienced engineers back into a quality process it had tried to automate, and just topped J.D. Power for the first time since 2010. Ford calls it a training data problem. The post argues the deeper issue is what automation removed. Automate your quality inspection and you automate the verifier, the capacity to know whether the automation works at all. Capture the judgment first, then automate, and the plan holds. Reverse the order and you spend three years and more than a billion dollars buying that judgment back. Weigh in on LinkedIn…
Sensing Is More Than Measurement. An internal Amazon presentation, reported by the FT, showed an AI project running 860 percent over budget, $1.8 million, that never shipped. Every token it burned sat on a monthly invoice for five months, and nobody noticed. The company that runs the cloud everyone else buys AI on could not see its own AI bill. Measurement and sensing are different jobs. Amazon had the number. What it lacked was the step that compares the number to an expectation and routes it to someone who can act while acting is still cheap. Weigh in on LinkedIn…
Can Frontier AI Outdo MBAs? Three top business schools tested frontier AI on MBA case work. On the headline partial-credit score, the leading model reached about 88 percent. On the stricter test, a complete answer with every criterion the instructor required, performance fell under half. The post works out why that gap matters. A score is a comparison, and a comparison needs a standard. The benchmark supplied one. Your hardest decisions do not, so the model is drafting and a person still owns the call. Weigh in on LinkedIn…
Could a Machine Have Had Darwin’s Idea? A digression, built from a thread I started on X in 2020, asking whether a machine could ever make the inductive leap Darwin made. The honest answer now is a qualified yes. Machines can generate the hunch, the part the old account of science thought had no method. But generation got cheap and verification did not, because verification in science is reality, and reality takes as long as it takes. What Darwin had that the machines still lack is the judgment to know which hunch was worth years of his life, and the patience to test it. Weigh in on LinkedIn…
One thread runs through all four. Generation got cheap. Checking did not. Ford rebuilt the people who can tell the machine it is wrong. Amazon lost track of a cost its own invoices spelled out. The benchmark scored high where a standard existed and went quiet where one does not exist. The machines produce Darwin’s hunches by the thousand and still cannot tell which one is worth a life. The output is cheap now. Owning whether it is any good is the work.
Before you go
The favor again, because timing matters. SXSW community voting closes August 23. If the book’s argument has been useful to you, a vote helps carry it to a stage in Austin next March. Vote here.
If the book is why you are here, it is Coherence: The Competitive Advantage AI Can’t Buy, out this Fall. Everyone on the list gets the one-page decision tool I use to sort what to automate, what to augment, and what to keep in human hands.
And if you try to coherise something this week, tell me how it went. Those stories are where the ideas meet reality, which is the only test that counts.