Redesigning Around AI is Nothing New.

"The engineer moved away from the parallel row of machines. New arrangements improved production flow, made better use of floor space and increased efficiency." — Warren D. Devine Jr., *From Shafts to Wires*, 1983

I have given this advice myself, more than once. Stop bolting AI onto the process you already have, and redesign the work around it. I still think it's right. It also, as stated, doesn't tell anybody what to move, which is why the objection that comes back is a fair one: the workflow is the core system, plus the org chart, plus twenty years of accumulated process. So "redesign the work" arrives sounding a lot like "replace everything." I don't think it is. I think it's one decision, it has a name, and this is not the first time Industry has had trouble finding it.

Electricity Arrived in 1881. Factories Didn't See The Full Benefits Until 1920.

Before that, in a factory, a steam engine sat in the basement, turning a vertical shaft that ran up through the floors of the building, and at every floor that shaft turned horizontal line shafts running the length of the ceiling. Off the line shafts hung leather belts, dropping down to each machine. And the machines were bolted into parallel rows underneath. The layout of the building was not a decision anybody made about the workflow, but rather a reality of the power source.

Then, electricity showed up. Edison and Swan had the carbon-filament lamp in 1879, and central generating stations opened in New York and London in 1881. Factory owners saw the potential so they bought electricity, which meant pulling out the steam engine and dropping in one big electric motor. The motor turned the same vertical shaft, which turned the same line shafts, which dropped the same belts to the same machines standing in the same parallel rows as before. On paper, that was a completed electrification project.

By 1900, according to Paul David's account in the American Economic Review, electric motors were still under 5% of factory mechanical drive capacity. That share took another 20 years to reach roughly half, and that is when productivity growth finally responded. The conventional account puts the gap at something like four decades.

The Motor's Gift Was Permission to Move the Machines, Not More Power.

The intermediate step was group drive. Instead of one engine turning everything, motors drove small groups of machines, so a department could run without engaging the whole building. That was a genuine improvement, but it was still a shaft-and-belt factory, and the machines stayed where the belts could reach them.

Unit drive was the change that mattered. One motor per machine, no belts, no shafting, no column of transmission running up through the middle of the building. Once the shafts came out, the thing that had been fixing the floor plan came out with them. Devine's history of the period puts it about as plainly as it can be put: the engineer moved away from the parallel row of machines, and the new arrangements improved flow, made better use of the floor, and increased efficiency.

Machines could then sit in the order the work moved through them instead of the order the shaft imposed. Plants could go single-story, because there was no vertical shaft to stack floors around. Materials could then travel through a building rather than around it. And how people worked changed along with the layout, because the layout had been holding the old way in place.

The machines themselves did not get better. Nobody invented a superior lathe. What improved was the arrangement, which had been sitting there looking like a law of nature.

So, if you run a carrier or an MGA, what is your shaft?

In Underwriting, the Shaft Is the Book.

For a long time my answer would have been the policy admin system, and that answer is wrong. A submission arrives and it already belongs to somebody before anyone has read a word of it, because the agent who sent it belongs to an underwriter's book or territory. Appetite screening, enrichment, quoting, and referral then arrange themselves around that person's day, because that's where the work physically sits. I've spent about a decade running production AI inside a live carrier, and it took me a while to see that the assignment, not the software, is the shaft.

So we bolt AI onto the steps. Prefill the application, extract the loss runs, score the submission on the way in. Those gains are real and I'd take every one of them — I have taken them. They are also capped at the boundary of the step they sit in, because the queue being fed is still the one the shaft dictated. A faster steam engine is genuinely faster.

Today, a submission is assigned first and read second. Rearranged, it's read first — appetite screened, gaps filled, priced-ready — and the underwriter attaches at the point where judgment is actually required. In most flows I've watched, a large share of the clock between arrival and quote is the submission sitting still waiting for a person, not being worked by one.

This is where an underwriting leader usually tells me the thing work is arranged around isn't software at all. It's the accountability structure. Authority, licensing, audit trail and comp are regulated or contractual, so there is nothing to move. Fair enough. I'd concede every piece of it. Authority letters don't move — line, limit, hazard class and state, audited by reinsurers under treaty. The producer relationship doesn't move either; reroute an agent away from their underwriter and distribution leadership hears about it inside a week. And clearance keeps its real job, which was never routing. It's broker-of-record and duplicate-submission conflicts, deciding who controls the account. What moves is when the accountable human attaches, not who is allowed to bind.

A submission triage engine is a product and several companies sell decent ones. The reordering around it — sequence, attachment point, what you measure — is not a product, and that's what nobody can sell you. The core system stays the system of record and receives the same transactions it always did. The factory kept its building, and it kept buying power from the utility; what it took out was the shafting.

Go look at what happens to a submission in the hours after it arrives, before anyone has read it. Not the cycle time you report on the monthly deck, but the dead time inside it. Almost nobody reports that number, which is part of how it stays where it is.

So is any of this new? Not really. The technology worked and it was easy enough to install, and pretty much nothing about the arrangement around it moved anyway. What I don't know is whether we do the same thing. If you drop a better model into the same sequence — assigned first, read second — you get a faster steam engine, which is a real improvement and a bounded one, and I've shipped a few. If you're willing to move the point where the human attaches, and leave the authority letters and the producer relationships exactly where they sit, you might have the other thing. Either way, tell me whether the sequence is serving the work or the org chart — or tell me I've named the wrong thing, which I'd want to know.