For most of history, having an idea for software and having software were separated by years. You needed a degree, or money, or a team, usually all three. The idea sat in a notebook and died there, and that was normal.
That gap has collapsed. Not narrowed. Collapsed.
I teach students who build working applications in a single semester with no computer science degree. Not mockups. Things that run, that other people use.
I find that genuinely remarkable and I notice how rarely anyone talks about it, because the AI conversation is mostly about jobs disappearing rather than about what just became possible for people who never had access.
What actually changed
Not that AI writes code. That is the visible part and the less interesting one.
What changed is that the cost of not knowing something collapsed.
Building software used to require holding an enormous amount of specialized knowledge before you could produce anything at all. Syntax, tooling, deployment, all of it front-loaded, all of it required before the first useful thing existed. That barrier filtered out almost everyone, and it filtered on patience for that particular kind of knowledge rather than on having good ideas.
Now you can start with the problem, learn the necessary pieces as they come up, and have something running before you have learned most of what a traditional path would have front-loaded.
That is a different order of operations, and it lets a different kind of person build.
Who this is actually for
The people best positioned here aren't the ones who want to be developers.
They are the ones who know a problem intimately and have never had a way to solve it. The office manager who knows exactly which report wastes six hours a month. The volunteer coordinator who knows precisely where the sign-up process breaks. The teacher who knows what the gradebook should do.
That knowledge is the hard part. It can't be acquired quickly and it cannot be outsourced to someone who wasn't in the room. What used to be missing was the ability to turn it into a thing, and that's the part that just got cheap.
If you have been carrying around a specific frustration for years, you are closer to fixing it than you have ever been.
What has not changed
Being honest about the limits, because overselling this is its own kind of harm.
Knowing what to build is still hard. AI will help you build the wrong thing very efficiently. Deciding what's worth building, for whom, is entirely on you and is most of the difficulty.
Maintaining it is still real work. Getting something running is now easy. Keeping it running, handling the edge cases, fixing it when it breaks at an inconvenient moment, is the unglamorous majority of the effort and nobody demos that part.
Judgment is not downloadable. Whether a system handles someone's data responsibly, whether it degrades safely, whether it should exist at all. Those require a person with a view.
You still need to understand what you built. If you cannot explain roughly what your thing does and why, you don't have a tool. You have a liability with a nice interface.
The part I want to say to people who think this is not for them
The most common thing I hear isn't "I couldn't learn that." It is "I'm not a technical person."
I do not think that's a real category. It is a story, usually acquired at some point when something was explained badly, and it becomes load-bearing without ever being examined.
I didn't learn any of this in a classroom. I asked, repeatedly, mostly of AI, and built things badly until they worked. That is available to anyone willing to be bad at something for a while, which is a much lower bar than a degree and a much less comfortable one, which is why more people don't clear it.
The permission you're waiting for isn't coming, because there's nobody whose job it's to give it.
What I would actually do
Pick the smallest real annoyance you have. Not a business idea. An annoyance. Something that wastes twenty minutes a week and has for years.
Describe it to an AI and ask how you would build something to fix it. Then do the first step.
It will be worse than you wanted. Do it anyway. The value isn't the tool, it is finding out that the gap between your idea and a working thing is now measured in evenings rather than years.
Most people never find that out, and it's not because they couldn't.