Dayton, OH Correspondent — Task-level gains are real and the aggregate hasn’t appeared, and both of those have been true for two years now.

The manufacturing view is that the useful applications are unglamorous — inspection, scheduling, predictive maintenance — and they’re being adopted steadily without anyone calling it a revolution.

Five years is enough time for the aggregate to show if adoption spreads past early users. It’s also the window in which every previous general-purpose technology failed to show it, and then showed it a decade later.

Orange County, CA Correspondent — The capital committed is the thing I’d watch rather than the technology.

Capital spending on data centers and related infrastructure by the largest firms is running at a scale comparable to a national infrastructure program, funded from cash flow rather than from debt, against a return nobody has demonstrated.

That’s a bet. If the productivity appears, it was the correct one. If it doesn’t appear within about three years, the correction is severe and the technology continues regardless, because the capacity will exist and the cost will be sunk.

Novi, MI Correspondent — Tools that raise output get used and the next generation should produce visible gains.

What we’re finding is that the constraint isn’t the model. It’s the data — ours is in forty systems that don’t speak to each other, and cleaning it costs more than the licence.

That’s why adoption looks slow inside large firms. The technology is ready and the plumbing is thirty years old, and the plumbing is the productivity project nobody wants to fund.

Myrtle Beach, SC Correspondent — We’re seeing jumps in specific jobs and I’d expect that to widen.

The honest thing from inside marketing is that the output is more voluminous and not obviously better, and the market hasn’t sorted the difference yet.

Everyone is producing more content. Nobody has more attention to give it. That’s a productivity gain that cancels itself, and I don’t know how it shows up in a statistic.

Las Vegas, NV Correspondent — A slice rather than a miracle.

The easy paragraph drops out of the hour. Whether that becomes productivity depends on whether a manager redesigns the work or just holds the same meeting with a chatbot in it.

Net gain if firms keep the exception handler and kill the wasted draft. Net shrug if they only cut the junior.

Long Island, NY Correspondent — Measurable in firms that own a workflow and rewrite it. Not measurable as a national number in this window.

The professional evidence is more mixed than the enthusiasm. There have now been enough incidents of fabricated citations in court filings to produce standing orders in dozens of courts requiring disclosure and verification.

That’s the shape of the problem in a high-stakes field. The tool is genuinely faster and its errors are of a kind the discipline isn’t built to catch, so the verification cost eats a large share of the time saved.

Tyler, Texas Correspondent — The roof still needs a person. The cover letter doesn’t.

I expect a bump rather than a rapture, and a bump is enough to reprice a credential, which is fine.

The insurance market is still declining to underwrite deployment in most high-stakes uses, and that’s the actual brake. Until there’s a standard of care and a loss history, the adoption in the sectors where it would matter most stays slow.

Cheyenne, WY Correspondent — It writes page one faster. Signers still sign.

Train signers and you get a print. Train more page ones and you get a glut.

Five years will show which we chose.

Owatonna, MN Correspondent — I’d offer the historical base rate, which is discouraging and shouldn’t be dismissed.

Electrification took roughly forty years to show in productivity statistics. Computing took about thirty, and there was a famous decade of economists arguing it was visible everywhere except the numbers.

Both eventually appeared and both required the work to be reorganized around the technology rather than the technology being added to the existing work.

Nobody has reorganized anything yet. Five years is optimistic on that basis, and being wrong about the timing is not the same as being wrong about the direction.

Prescott Valley, AZ Correspondent — The education effect is arriving faster than the productivity effect and it deserves attention.

We now have a generation of students for whom the first draft is free, and assessment methods built entirely on the assumption that it isn’t.

Whatever this does to the economy, it has already done something to how people learn to think, and the institutions are two years behind and mostly pretending otherwise.

Sheffield, Jamaica Correspondent — I’d point at the displacement side because my region is unusually exposed and it will happen before any measured productivity does.

Business process outsourcing employs a great many people in the English-speaking Caribbean, and it exists because the work is language work done cheaply in a compatible time zone.

That’s precisely the category. If the productivity gain arrives anywhere first, it arrives as the elimination of a national export sector here, several years before it appears in an American statistic.

The gain and the loss are the same event viewed from two countries.

Jacksonville, FL Correspondent — Ports are automating on a slower clock and the reason is labor rather than technology.

The productivity question in my industry was settled by a contract, not by a model. The port agreement this autumn restricted automation, and that’s a negotiated decision to forgo a measurable gain in exchange for employment.

Whether that’s right is a separate argument. What it demonstrates is that the aggregate number will be determined substantially by bargaining rather than by capability, and no forecast accounts for that.