More people are building software than ever.
Developers are building more because AI agents remove weeks of implementation work. Designers and product people can build working prototypes without waiting for an engineering team. People with little programming experience can make small tools for themselves.
This raises an uncomfortable question.
If anyone can build software, what is software worth?
The code was never the whole value
For a long time, code was expensive.
You needed years of experience, a team, a budget, and months of work. The difficulty of producing software made the finished product scarce. That scarcity became part of its value.
AI is removing some of that scarcity.
A feature that took a team two weeks might now take one developer two days. A small app that would have cost $20,000 to build might be created over a weekend.
That part of software is becoming cheaper. I don't think there's any point pretending otherwise.
But code was never the only reason software had value.
People pay for software because it solves a problem. They pay because they trust it with their data, because it fits their workflow, because other people they work with use it, and because someone will keep it running.
Generating the code is only one piece of that.
Cheap software creates more software
When something becomes cheaper to produce, we usually make more of it.
Digital cameras did not make photos disappear. They made photographs abundant. Website builders did not make websites worthless. They made it normal for every restaurant, plumber, school, and local club to have one.
AI will do the same to software.
We'll have software for problems that were too small to justify software before. A tool used by three people inside a company. An app for one family. A dashboard for a local sports club. A script with a nice interface that saves someone twenty minutes every Friday.
Most of this software will never become a business. It will still be valuable to the people using it.
I think this is one of the biggest changes ahead. We tend to discuss AI coding in terms of startups and developer jobs. The quieter change is that millions of tiny problems can now have their own tiny software.
Generic software will have a harder time
There is another side to this.
If your product is a thin layer around a simple workflow, people may decide to build their own version. Or a competitor may reproduce the useful parts in a week.
The code cannot protect you when code is cheap.
This does not mean every customer will build everything themselves. Most people do not want to maintain software, even if an agent can create it. I can cook dinner, but I still go to restaurants.
Convenience has value. Reliability has value. Support has value.
But the price of a generic tool will be pushed down when customers have more alternatives and competitors can enter quickly.
The small SaaS products that survive will need to be unusually good at a specific job. They will understand a customer better, fit into an existing workflow, own useful data, have an audience, or earn enough trust that switching feels risky.
"We built the software" is no longer much of an advantage.
The valuable part moves closer to the problem
As implementation gets cheaper, understanding becomes more valuable.
Knowing what to build is harder than producing another dashboard. Knowing which details matter to a dentist, a warehouse manager, or a teacher requires time around those people. An agent cannot recover context you never had.
Distribution matters more too.
You can build a good product in a week and still have nobody use it. When thousands of new products appear every day, getting attention becomes harder. An audience, a reputation, or direct access to customers may be worth more than the codebase.
Then there is operation.
The first version is cheap. Keeping a product useful for five years is not. Requirements change. APIs break. Customers need help. Security problems appear. Someone has to make decisions and take responsibility.
People will keep paying for that responsibility.
What this means for builders
I used to look at software and think, "That would take six months to build."
That thought stopped many ideas before they started. Now I often think, "I could make a version of that this week."
This is exciting. It is also dangerous, because building is a very convincing form of procrastination. You can produce a lot of software without creating much value.
The question I try to ask now is not whether I can build it. Of course I can, or at least I can try.
I ask who needs it, why they would choose mine, how they will find it, and whether I want to keep caring for it after the fun first week.
Those questions were always important. Expensive implementation let us avoid them for longer.
Software becomes a material
I think software is becoming less like a finished artifact and more like a material.
We will shape it around a problem, use it for as long as it helps, and replace it when the situation changes. Some software will serve millions of people. Some will serve one person for an afternoon.
Its value will come from what it makes possible, not how difficult it was to produce.
That is bad news if the only thing you sell is the difficulty of writing code.
It is very good news if you care about solving problems.