The Model Makes a Promise

On today’s flagship AI model release and what it does—and doesn’t—change

The demo reel is built for awe: code explained, sketches turned into sites, exams passed with scores nobody expected a machine to reach this quickly. The promise is that friction will melt—that what used to take an afternoon will take a minute, that the tedious layer between idea and execution can be stripped out by a polite paragraph to a patient system.

I am not immune to good tools. I like anything that shortens the distance between thought and draft. But the culture that forms around a tool matters more than the tool. If the new model becomes a way to publish faster without checking, we’ll get more of what already burdens the public square: confident wrongness, smoothed by grammar and multiplied by speed.

The core problem is assignment of responsibility. When a person says a false thing, we can ask them to correct it, and if they won’t, we can file them under “not credible.” When a model says a false thing, the company says “we told you it sometimes does that,” the user says “I was only experimenting,” and the false thing gets to keep walking. The harm doesn’t care who meant well.

Some changes are real. The ceiling for individual capability moves. A solo operator can now draft a competent memo, mock a landing page, and analyze a spreadsheet without hiring three different people. That’s not theft of labor; it’s a re-sorting of it. Specialists who bring context, accuracy, and judgment will still matter. The market for “pretty good and probably wrong” will be crowded and cheap. The market for “correct, accountable, and on time” will not go out of style.

Treat the model like a calculator that sometimes invents numbers. You don’t ban calculators; you teach estimation, show your work, and require answers to survive contact with reality. In writing, that means sources and the willingness to name them. In code, it means tests you run before you ship. In policy, it means decisions that do not outsource risk to a paragraph generator and then act surprised when the paragraph lies with charm.

The sensible metric for today’s leap is not cleverness in a demo; it’s what the tool lets a serious person accomplish with fewer unforced errors. A long time ago we decided spell-check did not make novelists obsolete. The same will be true here. Talent and discipline are not threatened by assistance. They’re exposed by it.