Agents are about to do years of work in days. Your dealership is still going to sell the same number of cars.

That’s the whole article. The rest is proof.

I’ve sat with well over 200 dealer groups this year. Almost every one asks the same question: How will AI help us sell more cars? Wrong question. Nobody buys a car every year because your BDC got faster. The right question is what happens to a business when the cost of getting work done falls toward zero and demand doesn’t move an inch.

History has an answer. It starts on a farm.

The long fall

In 1800, three out of four Americans worked a farm. Today it’s under 2%, and food output went up the whole way.

In 1850, one farmer fed about 4 people. By 1940, 19. Today, about 155. Same job title, 40 times the output. That’s what compounding a capability looks like.

Now notice what didn’t change. Americans still eat three meals a day. Farming hit a ceiling set by the human stomach. Remember that. We’ll come back to it.

34 years versus 3 years

The tractor took 34 years to beat the horse. Generative AI took 3 years to reach half the planet.

Horses and mules on American farms peaked above 25 million around 1920. Tractors didn’t outnumber them until 1954. Farmers needed financing, fuel, roads, mechanics, and dealers. (Yes, dealers. We’ve been here before.) Physical change moves at the speed of physical things.

Generative AI reached about 53% population adoption within three years, per Stanford’s 2026 AI Index. No new roads. No financing. It rode in on the phone already in your pocket. Your people are already using it. The question is whether your dealership is.

The clock breaks

Here’s where the analogy breaks. The tractor made a farmer stronger. Somebody still had to drive it.

Every machine of the industrial age multiplied what one worker could do in an hour. Output still scaled with worker-hours. The plant ran three shifts, not four, and it closed on Christmas.

Agents don’t need a driver for every hour. They work at 2 a.m., on Sunday, and sixteen at a time on the same problem. For the first time, the amount of knowledge work a business gets done isn’t capped by payroll or the clock.

That is the difference. The industrial revolution compounded muscle. The agentic revolution compounds the clock.

The doubling curve

The length of work an AI agent can finish on its own is doubling roughly every four months.

METR, an independent research group, measures how long a task an agent can finish on its own, in skilled human time. In March 2023, the best model handled a four-minute task about half the time. By early 2026: at least 16 hours. The doubling time is roughly 131 days.

Even at a stricter 80% reliability bar, that’s about three hours of expert work done unattended. And it compounds.

Now run them in parallel. In February 2026, an Anthropic researcher had 16 agents build a working C compiler from scratch: two weeks, about 100,000 lines of code, roughly $20,000 in compute. Sixteen agents, 24 hours a day, for 14 days is about 5,400 agent-hours. That’s two and a half years of one person’s working time, delivered in two calendar weeks. A human still steered. The human just stopped typing.

Software went first

A product idea that took a team of developers six to eight months can now become a working version in an afternoon with one person and good prompts.

Temporal’s August 2026 survey of 550 engineers found 51.3% now go from AI prototype to production-ready code in hours or faster.

Now the stat that should scare every dealer principal. DX studied 121,000 developers across 450 companies. Nearly all of them use AI. Organization-wide productivity is up about 10%, and it’s stuck there.

Same tools, wildly different results. Individuals are collapsing months into hours. Big organizations are bolting AI onto old workflows and getting a rounding error. We’ve seen this movie. It came out in 1890.

Bolting on the motor

Electricity reached American factories in the 1880s. Productivity did not take off until the 1920s. Forty years.

The late Stanford economist Paul David explained why. Factory owners pulled out the steam engine and bolted a big electric motor in its place. Same building. Same belts and shafts. Same layout. They got cleaner power and almost no productivity. The payoff came only when they redesigned the factory around the motor: a motor on every machine, the floor laid out around the work. Once they did, trend productivity growth jumped from 1.5% a year (1899 to 1914) to 5.1% (1919 to 1929). Same technology in both periods.

We’re doing it again. A 2026 NBER study of nearly 6,000 executives found 69% of businesses actively using AI and 89% reporting no labor productivity impact over three years. That’s a lot of electric motors bolted onto steam-era factories, and a lot of executives telling their boards they have an AI strategy.

In my whitepaper, Every Job in the Dealership, I model three scenarios for a franchised rooftop: Bolt On, Rewired, and AI Native. Bolt On is the dynamo in the steam factory. Rewired is the motor on every machine. AI Native is the factory designed for the motor from day one. The technology is identical across all three. The results are not.

The ceiling

Now the part nobody in your 20 Group is talking about. Agents make capacity nearly unlimited. They don’t make customers.

Cox Automotive expects 16.1 million new-vehicle sales in 2026 and 20.5 million used retail. That’s the size of the pie, and agents don’t bake a bigger one. Americans are keeping cars longer, not shorter: the average vehicle on the road is a record 12.8 years old. Nobody buys a second F-150 this year because your BDC answered a text at 2 a.m.

Service has a hard ceiling too, and it is made of people. Humans still fix cars. The US has about 805,600 automotive technicians, and TechForce’s 2026 report says new supply covers only 59% of annual demand, roughly 20,780 technicians short every year. An agent can book a thousand appointments tonight. It cannot turn a wrench tomorrow.

When capacity goes to infinity and demand stays flat, the value has to go somewhere. It goes to four places:

—Cost. SG&A is the first line agents attack. The winners won’t sell more cars. They’ll sell the same cars with far less overhead.

—Cash. Cox put new-vehicle days’ supply at 76 days in May 2026. On a unit near the $49,307 average listing price at a 7% all-in flooring rate, that’s about $700 of interest before anyone signs a buyer’s order. Every day agents pull out of recon, aged inventory, and contracts in transit is cash back in the bank.

—The bay. The technician is the constraint. Every minute your A tech waits on a parts pull, an approval call, or a write-up is capacity you can’t hire back. Agents should feed the bay so the wrench never stops. The play is more dollars per RO, not more ROs.

—Share. The industry’s demand is capped. Yours isn’t. Car retail is zero-sum: every unit you sell comes out of the store down the street. Speed and follow-up win that fight, and agents never forget to follow up.

Here’s the uncomfortable part. When every dealer has agents, the savings get competed away. The efficiency dividend is real, but it’s a race. First movers bank it. Everyone else hands it to the customer.

Who’s winning

The dealers who understand the ceiling are already moving. They aren’t the ones with the most AI tools.

Cox Automotive’s August 2026 tracker found 82% of dealers say they use AI. Most of it is busywork: task automation, follow-up, content. Many aren’t measuring it at all. Meanwhile, 63% of shoppers plan to use AI on their next purchase, and only 29% of dealers have started adjusting to AI-powered search. The customer is moving faster than the store.

The number for every GM’s desk: dealers working with an outside AI partner reported sales or revenue growth 36% of the time, versus 21% for those going it alone. In a capped market, that isn’t new demand. It’s share moving to the faster operator.

The agents are already doing real work:

—F&I. Tekion launched an F&I Manager AI in June 2026 that reviews a deal before it leaves the desk, catching missing documents and compliance gaps that delay funding.

—Service. A missed service call at 8:30 p.m. becomes a booked appointment in minutes, pulled from the store’s own capacity and op codes.

—Fixed ops revenue. Lithia says UK stores running AI service booking get 10% to 11% additional pre-sales on every AI-booked appointment. Before AI, close to zero.

—Back office. Baker Tilly reports AP invoice entry dropping from 10 to 12 minutes per invoice to as little as 15 seconds.

Lithia is the one to watch, because it isn’t bolting on. It’s redesigning the factory: almost 2.5 stores per office manager, about 1.4 stores per general manager, and a stated goal of SG&A under 60% of gross by 2028. Read that again. That’s not a sales strategy. It’s a cost and cash strategy.

One warning. An August 2026 review found no independent benchmark of any commercial dealership AI product. Most of the “hours to minutes” claims you hear at NADA come from the people selling the software. Get your own before-and-after.

Who loses

Every shift like this has losers. This one has three.

The bolt-on group. Five point solutions, five logins, no redesign. Same org chart, same headcount, bigger software bill. They’ll tell their board they have an AI strategy. What they have is an AI expense.

The headcount GM. If your status in the building comes from how many people report to you, the next 18 months will be uncomfortable. The scoreboard is moving to gross per employee and cash per unit.

The legacy vendor. Agents run on clean, open data. Any vendor that charges per seat, sits on the dealer’s data, and bolts a chatbot onto a decades-old system is selling the steam engine. Dealers will route around them.

The biggest winner may not be a dealer at all. It’s the customer, who gets faster answers now and, as the savings get competed away, better prices later.

Three calls for the next 18 months

Nobody will claim AI sold more cars. By spring 2028, the AI story on public dealer group earnings calls will be SG&A and cash, not units.

Stores per manager becomes the stat people brag about. What Lithia is doing today becomes the benchmark for every top group.

Fixed ops wins the AI race. Because the technician is the ceiling, the biggest measurable AI gains will show up in dollars per RO, not in the showroom.

What to do Monday

The tractor gave farmers a generation to adjust. Dealers will not get one.

If the doubling curve holds, agents by spring 2028 handle work 16 times longer than they do today. The tasks that eat a manager’s week, like month-end close, warranty claims, aged inventory decisions, contracts in transit, and the service follow-up nobody makes, become minutes of agent time. Run in parallel. Overnight. On Sunday.

That is the compounding no one is ready for. Not faster workers. Work that no longer waits for a worker.

So here is my challenge for every dealer principal reading this. Pick one department. Write down every task it does in a week and how long each one takes. Then ask two questions. If this work did not have to wait for a person’s hours, how would I design this department from scratch? And how many days of cash and how many tech hours does that give back?

That answer is your factory redesign. Everything else is bolting a motor onto a steam engine.

The farmers who sold their horses early didn’t grow more food than the country could eat. They bought their neighbors’ farms.

Where would you point agents first: cost, cash, the bay, or share?

Todd Smith is CEO of QoreAI, authored, The Intelligent Dealership: How AI and Data Transform Automotive Retail, and is among the scheduled speakers at Used Car Week that begins on Monday in San Diego.