The Subtitle I Almost Got Wrong

Newsletter – Edition 2

Those of you following my book’s journey from early on may have noticed that the subtitle has changed, and the reason is a small story about a mistake I almost made.

A few weeks ago my publisher told me, reasonably, that a reader likely couldn’t tell what the book was about at a glance. He wanted “agentic AI” in the subtitle, both for clarity and for search. At the time it read “What Wins When Everyone Can Go Fast,” which I liked and which never once said AI.

My instinct was to push back, and my first reason was not a good one. I didn’t want the book filed under yet another agentic-AI title when it’s a management book about the enterprise consequences of AI. That instinct had already cost me. Trying not to sound like an AI book, I’d written a subtitle that didn’t clearly signal what it was about.

The reason I actually cared about came out when I started drafting replacements. I didn’t want to compete for the noisiest keyword in the market. There are thousands of things shouting “agentic AI” right now, and being one more voice in that crowd is not the same as being found. I wanted words that would still hold up after AI stops being novel.

A quick test helped me sort the candidates. Say the subtitle out loud with “electricity” in place of “AI.” “The Last Advantage When Every Company Runs on Electricity” sounds a century old, which told me the phrasing was tied too tightly to the moment.

I settled on “Coherence: The Competitive Advantage AI Can’t Buy.” It names AI, which was the fair part of my publisher’s request. It makes an argument instead of chasing a search term, which was the part I wasn’t willing to give up. The advantage AI can’t buy is the one your competitor can’t buy either, because you’re both shopping from the same shelf. What isn’t on that shelf is the whole point.

The week in ideas

Three posts from the past week.

Good AI Governance Is Not the Same as Coherence. Australia’s directors just got one of the best AI governance guides I’ve read, and I spent the post explaining why the best version of the mainstream answer still misses the failure that will catch these boards. A governance apparatus works by review, and it reviews what reaches it. The incoherence that builds up between separately approved systems never comes up for a vote. And the human oversight everyone prescribes can pass its own audit while quietly going hollow, as a stretched review team waves through more and catches less. Real news from last week makes the point. Anthropic, the company that sells agentic AI, published a sober four-question checklist for deploying it safely: what untrusted content does the agent ingest, what can it do, what’s the blast radius, can you see what it’s doing. Four good questions. Every one of them inspects a single agent, one at a time. None of them can see the incoherence that accumulates in the space between agents that each passed. [Weigh in on LinkedIn…]

The Moat Is Coherence. Kirkland & Ellis, the highest-grossing law firm in the world, is spending around half a billion dollars to build its own AI rather than rent what its rivals can rent. Its chairman put the logic in a line: widely available tools raise the floor for everyone, and the firm doesn’t get hired for the floor. The post works out what Kirkland is actually buying, which isn’t the model and isn’t the data, but the coherence that turns both into judgment a competitor can’t copy. The giveaway is the exclusivity clause. If the value were the technology, keeping it exclusive wouldn’t matter, because the technology is for sale to everyone anyway. [Weigh in on LinkedIn…]

The Machine Proved It. Did It Do Mathematics? A digression, and my favorite of the three. An AI model recently disproved a conjecture the mathematician Paul Erdős posed in 1946, reaching across the field into tools no human specialist would have thought to try. The machine produced the proof. It did not choose the question. The post works through why understanding isn’t decoration but compression, the way a bounded human mind fits something enormous into the space of a single brain, and why the usual “trust the result you can’t follow” analogy from medicine breaks down in mathematics, which has no second way to verify a claim besides the proof itself. The closing point holds up under the whole argument. Nobody has built a system that decides which question is worth eighty years of human attention. [Weigh in on LinkedIn…]

There’s one thread through all three. The machine can produce the output. A human still owns the part with no dashboard: choosing the question, holding the separate pieces together, judging whether the answer is any good. Kirkland is paying half a billion dollars to own that part. The governance guides keep prescribing a version of it that passes the audit and can’t see. The mathematicians are the ones saying out loud that they don’t yet know how to measure it. It’s the same thing the new subtitle names. The advantage isn’t the AI. It’s the coherence around it, and that isn’t for sale.

Before you go

One more item, because it belongs to the same story. SAP just closed its acquisition of Prior Labs and committed more than a billion euros to a lab that builds foundation models for structured data instead of text. The bet is that the untapped value in enterprise AI sits in the tables and databases a business actually runs on, not in another chatbot. I think that bet is right and incomplete in a familiar way. Point a powerful model at a fragmented, contradictory data estate and you get fast, confident reasoning over an incoherent picture, which is worse than slow reasoning, because the confidence hides the mess. Making the data coherent enough to trust is the half nobody can sell you.

If the book is why you’re here, it’s 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 a good share of my ideas come from.