There is a strange feeling when an agent builds in one afternoon what would have taken you two weeks.
Part of you is delighted. The other part looks at the result and thinks, "Can this really be worth much if it was that easy?"
We have spent decades using effort as a rough measure of value. A large software project required a large team, a long schedule, and a large budget. When the finished product arrived, the cost felt justified because everyone could see how much work went into it.
AI breaks that connection.
The work gets shorter. The result can stay just as useful.
Customers never bought the hours
Imagine software that saves an accountant ten hours every month.
If it took a year to build, it saves the accountant ten hours. If an agent helped build it in a week, it still saves the accountant ten hours.
The customer gets the same result.
This is obvious when we talk about other products. A photographer does not charge less because a great photo took one second to capture. A mechanic does not charge only for the five minutes spent replacing the part. You're paying for the result, plus all the judgment required to get there.
Software has always worked this way too, even when we pretended otherwise.
A company pays for payroll software because people must be paid correctly and on time. It pays for monitoring because an outage costs money. It pays for a booking system because empty appointments are expensive.
None of those values depend on how many hours someone spent typing the code.
The cost still matters
This does not mean cheaper production changes nothing.
When the cost of making something falls, more people can offer it. Competition increases. Features that once supported an entire company become checkboxes inside a larger product. Some software becomes free.
The customer may value a tool at $100 per month. If twenty nearly identical tools compete for that customer, none of them can assume they will collect the full $100.
So two things happen at once.
Software keeps producing real value for the person using it. The builder has a harder time capturing that value when alternatives are everywhere.
This is the part that matters for software businesses. AI improves your margins today, then helps ten competitors show up tomorrow.
Billing for effort gets awkward
The old agency model often turned time directly into money.
Estimate the project at three months. Assign two developers. Multiply the hours by a rate. Add some margin. Send the proposal.
What happens when the same team can finish in three weeks?
You can hide the speed and keep billing the old estimate for a while. That will not last. Clients will learn what these tools can do, and another agency will quote less.
Hourly billing becomes harder to defend when productivity changes this quickly.
Builders will need to price the problem, the risk, and the result. A system that prevents a company from losing $500,000 per year can be expensive even if the first version took a week. A beautiful internal tool that saves one employee ten minutes per month probably cannot.
The hard part is knowing the difference.
Faster building makes smaller markets viable
Lower costs also open markets that never made financial sense before.
Suppose a product costs $200,000 to build. You need a lot of customers before it becomes a reasonable business.
If the same first version costs $10,000, the market can be much smaller. A niche tool for a few hundred architects, translators, or independent music teachers might be enough.
This is where I think AI gets interesting.
We already have plenty of software for huge horizontal problems. We have email clients, project management tools, note apps, and CRMs. The next wave can go much deeper into narrow work that large software companies ignored.
Those products may never become unicorns. They can still become excellent small businesses.
The first version loses its mystique
People used to treat a working application as proof that something significant had happened.
It meant money had been raised, a team had been hired, and months of execution had gone well enough to produce a launch.
Now a working application proves much less.
It proves that someone could get a first version working. That's useful, but it says nothing about whether customers care, whether the product is reliable, or whether anyone will maintain it.
The first version is becoming cheap evidence.
The valuable evidence comes later. People return. They pay. They trust the product with important work. The product survives contact with thousands of strange edge cases. The team keeps improving it after the excitement wears off.
Time saved during development does not remove this work. It lets you reach it sooner.
Who gets the extra value?
At first, builders keep most of the gain.
You charge the same price, use fewer people, and finish faster. Your margin improves.
Then the market notices. Prices fall. Products include more. Customers expect fixes in hours instead of weeks. Some of the gain moves from the builder to the buyer.
This has happened with every major improvement in software development. Better languages, open source libraries, cloud hosting, and app stores all made software cheaper to produce. We responded by building much more ambitious software.
AI is a larger jump, but I expect the pattern to rhyme.
A better measure
I don't want to measure software by how painful it was to make.
I want to measure it by the useful change it creates. Did it save time? Did it reduce mistakes? Did it let someone do work they could not do before? Does it keep doing that reliably?
AI can shrink the path between an idea and a working product.
That makes the path less valuable.
The destination still matters.