A teacher wants a tiny app to organize homework. A shop owner wants one that tracks unusual orders. They describe it in a chat window, watch an interface appear and get something that works. Until recently, most ideas this small would never have justified hiring a software team. But what happens to those apps after the first successful demo?
That question sits behind a striking claim making the rounds again this week: software may be approaching its “YouTube moment.” In a January essay, Andreessen Horowitz investor Anish Acharya argued that AI tools are shrinking the distance between an idea and a working app, opening the door to software built for tiny audiences or even a single person. A September 27 recap called this a “programming singularity.” The phrase is a prediction, not a measured milestone that experts have confirmed.
What the new tools actually change
The practical shift is real. Replit, for example, describes a workflow in which someone can start with a conversation, build an app and publish it from the same environment. A person who once needed to learn a framework just to test an idea can now ask for a prototype and revise it by describing what feels wrong. That creates room for the small, highly specific apps Acharya imagines.
But “can build an app” covers several different stages. A prototype that works for its creator on Tuesday is one thing. A service that stores other people’s information, survives a dependency update and still behaves correctly months later is another. The first result tells us the barrier to making software is falling. It does not settle the cost of keeping that software dependable.
Does AI make developers faster? It depends what they are doing
Research offers reasons for both optimism and caution. In three randomized field experiments spanning 4,867 developers, researchers estimated 26.08% more completed tasks for those given access to a coding assistant. That study examined assistance with code completion in company workflows; it did not test whether non-programmers could independently maintain production apps.
Another experiment found a very different outcome in a narrower setting. METR studied 16 experienced contributors working on 246 tasks in open-source repositories they knew well. With early-2025 AI tools allowed, they took 19% longer. METR explicitly warned against treating that result as a verdict on most software work. And in a February 2026 follow-up, it said its newer data could not reliably measure the current effect because developers and tasks were increasingly selected out when working without AI. The two studies measure different work, with different tools and people; their percentages cannot be added or directly compared.
The part a prompt cannot sign off on
Developer Simon Willison draws a useful line between generating code without reviewing it and using AI while reading, testing and understanding the result. This is a distinction about responsibility, not a rule that only professional programmers may build things. A one-person experiment can be valuable even if it will never have a support desk. Once an app takes payments, handles private data or serves other people, somebody must be able to explain what it does and fix it when it fails.
Why it matters
Cheaper creation could unlock software that never made economic sense before: a tool for one workshop, one classroom or one odd workflow. It may also produce many apps whose makers did not plan for the second year. Acharya’s comparison with YouTube is most useful as a question about who gets to create. Whether this becomes a durable software ecosystem depends on a less photogenic answer: who takes care of everything people build.
