Different surveys put the non-developer share somewhere between 63% and 84%, and the money behind it is real.
Two things are happening at the same time, and most articles pick one and ignore the other.
The first: hiring for entry-level software jobs has fallen sharply. Depending on which tracker you read, junior developer hiring is down somewhere between 25% and 67% from its 2022 peak. The estimates vary a lot because everyone counts differently, so treat the exact figure with suspicion. The direction is not in question.
The second: more people are building apps than ever before. On AI coding platforms, most users are not developers at all. Different surveys put the non-developer share somewhere between 63% and 84%, and the money behind it is real. Lovable was generating around $400 million in annualised revenue as of February 2026, roughly four times what it made eight months earlier. Bolt.new reached $40 million in annual recurring revenue within five months of launching.
So the job is shrinking and the craft is spreading.
Those are not opposites. They are the same event seen from two sides.
What actually happened to the junior role
Think about what a junior developer used to do in their first year.
Boilerplate code. Simple bug fixes. Writing tests. Drafting documentation. Building basic CRUD screens, meaning the standard create, read, update and delete parts of an app that every product needs and nobody enjoys writing.
Those are exactly the tasks AI tools now handle at a quality companies will accept in production.
The job was not deleted. The training ground under it was.
That creates a problem people in the industry have started calling the hollowed-out ladder. Plenty of senior engineers at the top, AI doing the routine work at the bottom, and very few people in the middle learning how any of it fits together. If it continues, the companies doing it will find in five years that they cannot hire seniors, because nobody made any.
Meanwhile the bar for the roles that do exist has moved up hard. Companies now expect a first-year hire to arrive already fluent with AI tools and able to ship a complete feature, not just write an isolated function. Skills that used to be taught on the job are now entry requirements.
What replaced it
Here is the part that gets less coverage.
The person who understands a problem deeply can now build the solution themselves. A product manager who knows exactly why a workflow is broken. A marketer who needs an internal dashboard nobody will ever prioritise. A domain expert in insurance, or logistics, or clinical work, who has watched developers misunderstand their field for a decade.
None of them had to wait for an engineering ticket this year. Some of them shipped in days.
That is a genuine expansion of who gets to build software, and it is bigger than the shrinkage in job postings. It is just spread across people who do not have “engineer” in their title, so it does not show up in hiring statistics.
Where it breaks
This is where most articles on the topic stop, and stopping there would be dishonest.
One analysis found that around 65% of vibe-coded apps carry security issues. AI-generated code can contain subtle bugs, security holes and performance problems that a non-technical builder will not spot. Anything handling payments or personal data without a review is a liability, not a product.
There is also a gap between working and durable. Tools that make the first version effortless can become brittle and expensive as the project grows. Plenty of builders have shipped in a weekend and then spent six months fighting the thing they shipped.
The practical pattern that seems to work: use AI to get from zero to one, prove people want it, then bring in real engineering to harden and scale what you have validated. Speed first, then strength. Not one instead of the other.
What the career actually looks like now
The skill did not disappear. It moved.
Writing syntax used to be the hard part. Now the hard parts are knowing what good software looks like, being able to read and judge code you did not write, understanding what breaks under load, and having the taste to reject an AI’s first answer.
Hiring is starting to reflect that. Some teams now ask candidates to build something live with AI assistance during the interview and watch how they work. That tests something a whiteboard never did: can you direct a tool, spot when it is wrong, and recover.
It is also worth knowing who is still hiring at entry level, because it is not nobody. AI tooling companies and their ecosystem are actively hiring juniors, since they need people who understand the stack deeply. Amazon has said it still plans to take 11,000 interns in 2026. And companies running systems that must live for twenty years keep hiring juniors regardless, because they measure reliability and knowledge transfer rather than tickets closed.
Four things worth doing
Ship something real and keep it running. A live project with actual users, monitored uptime and a written note about the time it broke tells a hiring manager more than a portfolio of finished-looking screenshots. The running part is the signal, not the launch.
Learn to review, not just to generate. Get comfortable reading code you did not write and asking what happens when it fails. That is the durable skill.
Pick a domain you already understand. The advantage in this market is not coding speed. It is knowing which problem is worth solving, which is knowledge tools cannot supply.
If you are early in a career, chase mentorship over title. A team with senior people who review your work is worth more right now than a better-sounding role at a company that has stopped growing anyone.
The uncomfortable summary is that app building became easier to start and harder to be good at. Both halves are true, and only one of them is in the headlines.
