In the mid-1970s, GEICO was a few months from running out of money. The stock had fallen from around sixty dollars to about two. The company had spent years writing more policies than it could afford to pay out, which is a pleasant way to grow right up until the bills arrive. A new chief executive named Jack Byrne walked in and did the least celebrated thing a leader can do: he made the company smaller. He raised prices, dropped customers, and pulled out of whole states. He fixed the thing underneath before he touched the thing on top. The growth everyone wanted came later, and only because he refused to chase it first.
I keep that story close this week, because the loudest signal in the market right now is the opposite instinct. There is a version of every company, every product, and every technology that looks ready, and there is a version that is ready, and the distance between them is where nearly all the real work lives and almost none of the attention does. The market pays for the first version. The second version is the only one that survives contact with a customer, a claim, or a Tuesday afternoon when nobody is watching.
Why “Not Yet” Is the Most Honest Answer
The headline doing the rounds is that the most-watched company in artificial intelligence has not held a single pre-IPO investor meeting and has not set a timeline for going public. CNBC’s reporting on that absence was careful about what it did and did not establish, sourcing the claim that OpenAI has not held pre-IPO investor meetings to people familiar with the matter rather than to any announcement. Read one way, that is a non-event, a company declining to do a thing it was never obligated to do. Read another way, it is the most interesting thing the company has said all quarter. The entire market is leaning forward asking for a date, and the answer is a shrug. No meetings. No timeline. Not yet.
The market hates “not yet.” A date is a story you can trade; you can build a position around it, write a note about it, and be early to it. “Not yet” is just the truth, and the truth doesn’t move quickly enough to be exciting. Notice what the absence of a timeline actually tells you, though. It tells you nobody inside the building has decided the thing is finished. That is information, and it is better information than a date would be, because a date is a promise about the future while “not yet” is a report about the present. One of those is checkable.
The Demo Is Not the Deployment
Somewhere in the same week, there’s a quiet recurring conversation, a weekly one, about bringing agentic AI to production. Notice that it has to be a recurring conversation. Nobody schedules a standing weekly meeting about a solved problem; standing meetings are what an organization creates when a question keeps coming back and refuses to close. The phrase “to production” is doing enormous work there. It is the whole distance between a thing that demos beautifully and a thing you can actually leave running while you sleep.
A demo is a performance of readiness. Production is the absence of an audience. One is built to be watched, and the other is built to be forgotten about, which is much harder, because forgetting about something is the highest compliment you can pay it. It means it works. It means the failure modes have been found and handled rather than routed around during rehearsal, that the odd input got a real answer instead of a shrug, and that nobody has to be standing by. A demo can be made to succeed by controlling the conditions. Production is defined by the fact that you no longer control the conditions.
This is the same fault line running under the IPO non-news and the GEICO story. Byrne’s company looked like a growth business right up until the claims came due, because writing a policy is a demo and paying a claim is production. The policy is the version that looks ready. The claim is the version that tests whether it was. Every industry has its own version of this pair, and the interval between them is the only honest measure of how much a company actually knows about itself.
Scarcity Is a Feature of Value
A fourth signal floated past me phrased as a question: how do you ration a consumer right? It reads like a finance puzzle, and it is, but underneath it is something simpler. The moment something is worth having, you have to decide who gets it and when. Rationing is just the honest admission that not everything can be available to everyone at once, and that pretending otherwise is a decision too, usually a worse one.
A company that won’t name an IPO date is rationing certainty. A team that keeps an agent in testing instead of shipping it is rationing trust, releasing it only when it has earned the release. Byrne rationed GEICO’s own growth, deliberately, when growth was the only thing the market wanted to buy. In each case the restraint looks like weakness to the person who wants the thing now, and it looks like discipline to the person who has to live with the thing later. The difference between those two readings is almost entirely a difference in time horizon, which is why the argument never resolves: the two sides are not disagreeing about the facts, they are standing at different distances from the consequences.
We have built an economy that treats availability as the goal, everything instantly to everyone, and that instinct is strong enough that withholding now reads as a failure of nerve rather than a judgment about quality. But the things that hold their value tend to be the ones that made you wait. Not as a marketing trick, and not as manufactured scarcity dressed up as craft. As a side effect of actually being finished.
Profit Is a By-Product, Not a Plan
There’s an idea I keep circling: that a great customer experience produces profit as a happy by-product, and that aiming straight at the profit puts you directly at war with the customer you’re supposed to serve. Aim at the experience and the money follows. Aim at the money and you start writing policies you can’t pay out, which is precisely what nearly killed GEICO. The mechanism is not mysterious. Every shortcut that improves this quarter’s number is a shortcut taken out of somebody’s future experience, and the bill arrives with interest and without warning.
The same logic scales up to the whole AI moment. Aim at the valuation, and you set a date, hold the meetings, ship the half-ready agent, and perform the inevitability. Aim at the thing actually working, with the underwriting sound, the agent durable, and the experience worth having, and the valuation becomes a by-product you don’t have to chant for. The companies that get this backwards are not stupid; they are responding rationally to a scoreboard that updates daily while the underlying quality updates over years.
The market keeps asking these companies when they’ll be ready and treating the absence of a date as a problem to be solved. It has the question backwards. The companies that survive their own hype are usually the ones disciplined enough to say “not yet” while everyone else is selling “now.” Byrne’s “not yet” saved a company everyone had already written off. We remember it as a turnaround. At the time it just looked like a man refusing to perform, and the people who called it weakness were reading the same facts as the people who called it discipline.
The hardest thing to sell is patience, because it has no launch date, no slide, and no moment at which anyone gets to applaud. But it is the only thing on this list that has ever reliably compounded. Everything else, the timeline, the demo, and the announcement, is just the appearance of readiness, rented by the day. Real readiness doesn’t announce itself. It just quietly keeps working after the room has emptied out.
Leave a Reply