An Audience of One
When the cost of building falls far enough, software can serve an audience of one — but judgment still has to come from somewhere.
The Indie Comeback
I saw a post not too long ago, suggesting that AI might power the comeback of the indie developers who shaped so much of the software industry during the 1990s and early 2000s.
The argument was that AI is lowering the barrier to creating software, which would allow individuals and small teams to build the kinds of focused, opinionated and quirky products that have become impossible to find as the industry has metastasized around large companies and platforms.
I think there's a lot of truth to this, but it got me thinking about something larger.
What happens when the cost of creating software falls far enough that audience size is no longer the deciding factor to the question "Should I build this"?
An Audience of One
For most of the history of software, an idea needed a sufficiently large market to justify the time, money, and experience required to create it. Even independent and open source developers had to make that calculation at some point. A product could serve a relatively small niche, but that niche had to be large enough to support the work.
AI has the potential to change that calculation entirely.
If we play our cards right, agentic systems could hurl us into a world where software can be designed for any audience size, including an audience of one.
Think about it:
- A parent building something specifically for their child
- A teacher creating a tool for one classroom
- A small business replacing a frustrating manual process that no software company would ever consider valuable enough to address
It would mean that people no longer have to rely on a company deciding that their problem is worth solving.
I find that possibility genuinely exciting.
Importance Over Opportunity
There are an infinite number of problems that have gone unsolved, not because they were unimportant to the people struggling with them, but because there weren't enough of those people to make solving the problem commercially viable. AI could make it possible to build software around the importance of the problem rather than the size of the opportunity.
That is a future I hope we reach, but it's also a future that contains a great deal of risk.
The Promise You're Making
When you produce software for someone else to use, you're making a promise, whether you know it or not. You're telling that person that the software has been designed with reasonable care and that it is safe enough to trust with whatever information, responsibility, or access it needs.
Most people using software can't evaluate whether that promise has been kept. They can't inspect the architecture, review the dependencies, examine the authentication system, or figure out if their data can be recovered after something goes wrong. They have to trust that the person who created the software understood the responsibility they were accepting.
That responsibility doesn't disappear because the software was easy to make, and this is where the promise of AI-assisted development becomes more complicated.
Judgment Still Matters
AI can help someone create working software without requiring them to spend years learning how software is built. That's part of what makes it so powerful. It allows people to move from an idea to something tangible without first acquiring the formal training or hard-won experience that the industry currently demands.
The problem though, is that the need for judgment hasn't disappeared with the need to write every line of code by hand.
A person can now create an application without understanding the risks introduced by its architecture, the packages it depends on, the way it stores information, or the permissions it requests. They can see that the application works and reasonably believe that working is the same thing as being ready for someone else to use.
It's not, at all.
What Experience Remembers
Those of us who have built software for a long, long time, tend to carry around a collection of lessons that are invisible from the outside. Much of what we call experience is the memory of systems that failed, assumptions that proved false, data that could not be recovered, and decisions that seemed harmless until they reached production and suddenly weren't.
We learned to ask certain questions because we remember what happened when nobody asked them.
Vibe coders and the agentic systems helping them build are not yet equipped with all of that judgment. The systems are becoming more capable at producing code, but producing code is only one part of producing software that can be trusted.
Build Anyway
I don't believe the answer is to tell people they should stop building. That's stupid.
I want this explosion of software creation to happen. I want people to build tools for their families, businesses, communities, and themselves. I want useful ideas to become real without first needing to satisfy a venture capitalist, attract a large market, or fit inside the roadmap of an existing software company.
On the other hand, I also want the people using those tools to be protected by more than the confidence of the person who prompted them into existence.
If AI is going to remove the experience requirement from creating software, then we need another way to provide the judgment that experience used to produce. We need systems that help builders recognize risks they don't know how to see, ask questions they don't know to ask, and verify claims that neither they nor the agents working for them should be trusted to make without evidence.
Making Judgment Available
That is the problem I have been working on. It began with a set of principles about how humans and agents should work together. Those principles led to reusable skills, operating practices, verification systems, and software designed to give agentic work a stronger foundation.
Eventually, that work became The Room.
The Room is not built around the assumption that people need a better collection of prompts. It's built around the belief that making software available to everyone also requires making sound judgment, verification, and responsible practice available to everyone.
Judgment, verification, practice
Principles, skills, and systems that help builders see risks they would otherwise miss — and verify claims before someone else has to trust them.
The Future Worth Building
I honestly believe AI could lead to the most creative period of software development we have seen in a long time. It could produce a world filled with strange, personal, useful applications created for audiences that were thought too small to matter before.
Some of those audiences may consist of a single person.
I want that future to exist, but I also want the software created within it to deserve the trust people place in it. That is the future I am trying to help usher in.