In 1881 the first central generating stations opened for business in New York and London, and the men who understood electricity began describing the factory it would make possible. They were right about almost everything. They were wrong about the timing by roughly forty years. In 1899, according to the economic historian Paul David, electric motors still accounted for less than 5 percent of the mechanical drive in American factories. It took about two more decades to reach half, and the great gains in manufacturing productivity did not arrive until the early 1920s. A contemporary, David wrote, might have remarked that the dynamos could be seen everywhere but in the productivity statistics.
The delay was not a failure of the technology. It was a failure of the floor.
The first electrified factories did the sensible thing. They kept the old building, with its steam engine below and its great line shafts running overhead, and added electric motors to turn sections of the shafting, an arrangement engineers called group drive. The original engines, belts and shafts remained in place. The multistory layout, built to limit the power lost in turning very long shafts, remained too, and the machines stayed exactly where the shafting required them to be. Owners paid for the new power and kept every constraint of the old one, and the productivity accountants, as David puts it, had to reckon the old equipment as capacity still standing on the books.
The Motor Beside the Engine
The payoff came only when engineers stopped treating the motor as a replacement for the engine and started treating it as a reason to redraw the building. David describes what the "unit drive" made possible once each machine had its own motor: lighter construction because nothing heavy hung from the ceiling, single-story plants laid out along the flow of materials, machines arranged around the work rather than around the shaft, and the ability to change one line without stopping the whole mill. None of that was visible in the motor. All of it was visible in the floor plan.
And the floor plan had to be learned. David is careful on this point. Wide adoption, he writes, required "building up a cadre of experienced factory architects and electrical engineers familiar with the new approach to manufacturing," a slow and uncertain process that depended on people working out the details plant by plant. The technology arrived in a crate. The redesign arrived in the heads of the people who had run the old floor and were willing to think hard about the new one.
I have been reading three pieces of research this month that describe marketing departments standing in that same old factory, with a new motor humming beside the old engine and a decision to make about the crew.
Three Readings of the Same Plant
The first comes from Forrester's 2027 predictions for business-to-business marketing, published last week. As pressure grows to reduce personnel costs with AI, the firm expects marketing leaders to "eliminate as many as 25% of roles in anticipation of productivity gains that fail to meet expectations." The sequel is the instructive part. Having cut, those leaders will have to choose between spending program budget to restore capacity through rehiring, contractors or agency support, or accepting weaker execution with a smaller team. Either way, Forrester concludes, marketing's contribution to the business shrinks.
The second is a survey of more than a thousand business software buyers in G2's 2026 AI Agents Report, which gives the gap between promise and payoff a name: the distrust tax. Thirty-one percent of buyers say employees spend six or more hours a week supervising agent work. Forty-five percent say the time spent overseeing and correcting agents consumes more than a quarter of the productivity gains they expected, and 46 percent report higher operating costs than they planned. Almost half of the organizations running agents in production still govern them through informal or decentralized arrangements.
The third is the fourth annual AI and the Future of Marketing report from Basis, a survey of 184 agency and brand professionals. In the past year, 47.8 percent of their organizations replaced human tasks with AI, up from 27.9 percent the year before. Just 6.5 percent evaluate their AI investments against defined metrics and benchmarks, and 65.1 percent either do not measure AI's return at all or do not know whether anyone does. Only 13 percent describe their first-party data as extensive and well structured.
Put those numbers in one room and the picture is a factory owner who has installed the motor, left the shafts in the ceiling, hired a man to watch the belts slip, and is now deciding how many machinists to send home on the strength of a productivity gain that nobody has measured.
The Crew Is the Cadre
The temptation is understandable, because the arithmetic of headcount is the only arithmetic that is easy. A salary is a number on a spreadsheet. A redesigned workflow is a set of decisions about who owns which judgment, which data the system reads, which steps can run without review and which must stop for a signature. The salary can be removed in a week. The workflow takes a quarter to draw and a year to settle, and it does not appear in any line of the budget until it starts paying.
So the cut comes first, and the floor stays as it was.
What gets lost in that order of operations is the cadre. The people most likely to be cut in a productivity program are the ones who know where the old belts ran: the coordinator who knows which approvals actually matter, the analyst who knows why the pipeline report and the finance report disagree, the producer who knows which brand rules are written down and which live only in memory. Those are precisely the people a redesign needs, because the new floor plan is drawn from knowledge of the old one. An agent cannot be told to skip a step that nobody remembers is there. It cannot route around a dependency that was never documented. The distrust tax G2 describes is, in large part, the price of running new machinery on a floor whose logic left the building with the people who carried it.
There is a second loss, quieter and more expensive. Forrester's prediction says the cut will be reversed through rehiring, contractors or agencies. That means paying, at a premium and in a hurry, for capacity the company already had, and then asking the new arrivals to learn a floor that is now stranger than the one the old crew understood. It is the factory owner who scrapped the line shaft before buying the motors, and is now renting a steam engine by the hour.
Draw the Plan, Then Staff It
None of this is an argument against change in the size or shape of a team. The electrified factory did employ people differently, and so will the marketing department that learns to work with agents. The argument is about sequence. The factories that won drew the new plan first and then staffed it. The factories that lost bought the motor, kept the shafts and blamed the electricity.
In practice the sequence is not mysterious. Before any role is cut on the promise of AI, the work that role does should be written down as a flow, with its inputs, its decisions and its handoffs, plainly enough that a new person or a machine could follow it. Each step should be sorted by risk, so that the reporting and drafting and checking can run on their own while anything that spends money, speaks publicly or changes a customer record waits for a person. The gain should be measured against a baseline taken before the change, in hours and cost per finished piece of work, so that the productivity case is an observation rather than a forecast. And the people who carry the old floor in their heads should be the ones asked to draw the new one, because nobody else can.
G2's own data contains the hopeful version of this story. Among teams running agents inside mature workflows, four in five report cycle time gains of at least 11 percent. That is not the tenfold promise. It is the modest, compounding return of a floor that has actually been rewired, and it belongs to the teams that did the redesign rather than the ones that skipped it.
The dynamo was never the factory. It was permission to build a different one, and the companies that profited were the ones that kept the people who knew how the old building worked long enough to design the new building around the work. Marketing has its motor now. The question for next year's plan is whether it will redraw the floor, or simply send home the only people who know where the shafts are.
