The agency model breaks when the agency is software

An agency sells you forty hours and delivers ten hours of judgment wrapped in thirty hours of typing. You pay for all forty, because the forty is the unit. The hour is the thing on the invoice, the thing on the timesheet, the thing the whole business is built to count. Then an agent does the thirty hours of typing in nine minutes, and the invoice has nothing left to count.

That’s not a productivity gain. That’s a hole where the business model used to be.

Here is the line this whole post orbits: the agency sells the hour with a markup, and when the work is done by an agent, the hour disappears. Everything that follows is what happens next, to the pricing, to the margin, and to the kind of company an agency has to become to still exist on the other side.

The hour was never the product. It was the proxy.

Start with the naive picture, because almost everyone inside an agency still lives in it. The product is talent. You hire skilled people, you rent them out by the hour, you mark the hour up, and the spread between what you pay them and what you charge the client is the business. Bill more hours, hire more people, grow. The model is so old it feels like physics.

It isn’t physics. It’s a proxy. Nobody buying a brand campaign or a tax return or a sprint of code actually wants hours. They want the campaign, the return, the working feature. The hour was just the only honest way anyone had to price uncertainty, you couldn’t promise the outcome, so you sold the effort, and the effort was measured in time because time was the one thing you could verify. The client trusted the clock because the clock was the only witness.

So the billable hour was never the value. It was the receipt for the value, in a world where the value itself couldn’t be counted directly. That distinction sat harmlessly under everything for decades, because the receipt and the value moved together. More hours, more output, more or less.

Then they came apart.

When the typing collapses, the markup collapses with it

Watch what an agent does to the ratio. Most billable work, and every agency knows this, even the ones who won’t say it, is not the rare expert judgment. It’s the execution around the judgment. The strategist decides what the campaign should say in an afternoon; the rest of the week is producing the variants, formatting the deck, drafting the emails, reconciling the comments, rebuilding the thing after the client’s third round of notes. The thinking is the small part. The typing is the billable part.

An agent eats the typing. Not all of it, not perfectly, but enough that the thirty hours becomes three, and the three are mostly the human checking the agent’s work. The judgment is still there. The judgment got more valuable, if anything. But the hours it used to drag behind it, the hours you actually invoiced, are gone.

Now run the old business model forward with that single change, and watch it eat itself.

You did the same job. You delivered the same outcome, arguably a better one, faster. And you can bill a fraction of what you used to, because you bill the hour and the hours are gone. The agency that gets more efficient under the hourly model gets poorer. Efficiency is punished. The better your agents, the smaller your invoice. That is the trap, stated plainly: the agency sells the hour with a markup, and when the work is done by an agent, the hour disappears, and the model that’s left rewards you for being slow.

The first reflex is to hide it. Keep billing the forty, let the agent do the work, pocket the difference quietly. That works for exactly one quarter, until the client’s own agent does the same job in the same nine minutes and the client realizes what they were paying for. The markup on labor only survives while the labor is scarce. The moment the client can summon the same labor, the spread is gone, and pretending otherwise just teaches them to leave.

You can’t fix a broken unit by negotiating the price of it

The honest reaction is to reprice. And the first repricing everyone reaches for is the wrong one, so it’s worth walking into the dead end on purpose.

The dead end is: keep the hour, just charge more per hour. “Our people are AI-augmented, so each hour is worth more.” It sounds reasonable. It dies on contact with the math. You can triple your rate, but if the work that took forty hours now takes four, you’ve tripled a number you multiply by a tenth. The client doesn’t see a premium hour. The client sees a four-hour invoice where a forty-hour one used to be, and concludes, correctly, that the work got cheap. You’ve repriced the unit without changing the unit, and the unit is the problem.

The second dead end is the subscription with no anchor: “pay us a flat monthly fee for unlimited AI-powered work.” Now you’ve removed the clock entirely, which sounds modern, but you’ve replaced it with nothing. The client has no idea what they’re buying. You have no idea what you’re promising. The flat fee is just an hourly model with the meter hidden, and the first month the client’s needs spike, one side feels robbed.

Both dead ends share a mistake. They treat this as a pricing problem, how much per unit, when it’s a unit problem. The thing you’re selling has to change, not the price tag on the old thing.

The unit that survives is the outcome. Not the hour spent, the result delivered. Not “forty hours of design,” but “the brand system, shipped, that converts.” Not “a sprint of engineering,” but “the feature, live in production, that does the job.” The moment you price the outcome instead of the effort, the agent stops being a threat to your invoice and becomes the engine of your margin, because now, doing the work faster makes you more money, not less. The agent’s speed accrues to you, the way a software company’s marginal cost accrues to the company, not the customer.

That sentence is the whole pivot, and it has a name. When your delivery is an agent and your pricing is an outcome, you are no longer an agency. You are a product.

The agency that survives becomes a product

Here is the shape of the thing on the other side, and it’s worth being precise, because “become a product” is the kind of advice that sounds like a poster and means nothing until you make it concrete.

A traditional agency scales by adding people. Twice the revenue needs roughly twice the headcount, because the deliverable is human hours and humans don’t fan out. Margins are thin and flat, the spread on labor is the spread on labor, and it doesn’t improve as you grow. The org chart is the product, and the org chart caps the company.

A product company scales differently. It builds the thing once and sells it many times. The marginal cost of the next sale is near zero, so margin expands with scale instead of staying flat. The asset isn’t the team’s time this month; it’s the accumulated thing the company has built, the system, the brain, the repeatable engine that turns an input into an outcome without a human re-doing it each time.

An agency whose delivery is an agent can finally be that second kind of company. The first time it solves a problem for a client, that solution doesn’t evaporate when the invoice is paid. It compounds. The agent that ran the campaign learns the playbook. The system that filed the return remembers the edge cases. The next client’s job starts from everything the last client’s job taught the machine, and the human expert moves up to the judgment that’s genuinely scarce, instead of down into the execution that no longer is.

This is why the winners won’t be the agencies that adopt agents fastest. Adoption is table stakes; everyone gets the same models. The winners will be the ones that change their unit, that stop selling the hour and start selling the outcome, and build the compounding system underneath that makes the outcome cheaper to deliver every single time. The agency that just bolts an agent onto an hourly invoice is automating its way to a smaller invoice. The agency that becomes a product is automating its way to a bigger one.

Say you run a small studio that does this. The work that once needed a room full of people now needs a handful of experts and a system that does the rest. Suppose a job that used to fill a week now ships in an afternoon. Under the old unit, that’s a catastrophe, you just lost four days of billing. Under the new unit, it’s leverage: the same outcome, delivered faster, at a price tied to the result, with the saved days spent winning the next client instead of grinding out the last one. Same agents. Opposite outcome. The only thing that changed was the unit.

The turn: the expert was never the bottleneck. The hour was.

Strip away the business model and there’s a person in the middle of all this, and they’ve been miscast for a long time.

The expert at an agency, the strategist, the senior engineer, the seasoned accountant, was always the valuable part. But the hourly model forced them to spend most of their day on the part that wasn’t valuable: the production, the formatting, the rounds, the typing. The thing the client actually paid a premium for, the judgment, was the thing they had the least time to do, because the unit they were sold by demanded they fill the hours with execution. The model didn’t just mis-price the work. It mis-spent the person.

When the agent takes the typing, the expert finally gets to be the expert full time. Not faster at producing variants, freed from producing them. The day stops being execution with judgment squeezed into the cracks, and becomes judgment with execution handled underneath. That’s not a smaller job. It’s the job they were hired for and rarely got to do.

So the real story here isn’t that agents threaten agencies. It’s that agents threaten a pricing model that was always slightly dishonest, one that sold effort because it couldn’t price outcomes, and trapped its best people in the effort to keep the meter running. Take the meter away and the question changes from “how many hours did you spend on me” to “what did you actually deliver.” That’s a better question. It’s the one clients always wanted to ask, and the one experts always wanted to be judged on.

---

That’s the bet we’re making at Apollo Space: that the company of the near future isn’t an agency with faster typists, but a product with experts on top, a compounding system that turns inputs into outcomes, with humans doing the one thing the machine can’t, which is deciding what “good” means. The agency model breaks when the agency is software. What replaces it is the agency that was honest enough to stop selling the hour, and brave enough to put its name on the result.