To cost you your job, AI only has to be cheaper than you.
I have been programming since 1984. In that time I have watched a fair number of technologies arrive that were going to change our work forever.
Some of them did.
We went from assembly and BASIC to higher-level languages. From hand-written HTML to frameworks. From our own servers to the cloud. From Stack Overflow to an agent that walks half your codebase before your coffee goes cold.
And at every step, the same thing was said: developers are about to become redundant.
It never happened.
This time something is different, though. AI has not suddenly become intelligent, and Claude and Codex still write software with bugs in it. The difference is economic.
You don't have to be replaced to lose your job
Say you have a team of twenty developers.
In two years, ten of them might ship the same amount of software. Those ten are the same developers, with agents taking over most of the execution.
Then the company has a choice.
Keep twenty people and build twice as much. Or keep ten and keep delivering the same.
Which one do you think the CFO is going to pick?
That is no longer a thought experiment. In February, Jack Dorsey's Block went from over 10,000 employees to fewer than 6,000. Dorsey gave the reason himself: "We're already seeing that the intelligence tools we're creating and using, paired with smaller and flatter teams, are enabling a new way of working which fundamentally changes what it means to build and run a company."
TechCrunch keeps a running list of the big tech companies that named AI in a round of layoffs this year. By the end of July it had 21 on it.
Of course this is a simplification. Software development is more than writing code. Requirements are vague, systems are full of historical decisions, and production has the annoying habit of ignoring your benchmarks.
Some of those layoffs are also plain cost-cutting with a more modern story attached.
But the argument that AI cannot replace a complete developer misses the point.
It doesn't need to.
Demand is growing again, just not for everyone
Since this summer, the counter-argument has numbers too.
According to Indeed Hiring Lab, US software development postings grew almost 15% after Claude Code launched in late February 2025, while total postings fell 7% over the same period. After years of contraction, the occupations most exposed to AI are now rebounding the hardest. Indeed is careful to say correlation is not causation, but calls the timing "a coincidence that cannot be ignored".
That is good news, and I am not going to explain it away.
Look at who is being hired, though. Of the growth between May 2025 and May 2026, 71% came from senior roles and 37% from jobs with AI in the title. Indeed's own reading: "demand is growing for experienced professionals who can work with AI, not necessarily a broad-based recovery across all software roles."
So the market is growing and shrinking at the same time. Block cuts its headcount, the job boards fill up again, and which side you land on depends on who you are.
The junior goes first
Every senior developer spent years fixing bugs, making the wrong architecture calls, digging through production incidents, and learning why that one elegant solution turned out to be a disaster in practice.
You don't learn that from a prompt.
But if an agent handles the simple tickets faster than a junior, why would a company still hire the junior? That 71% is the answer the market is giving right now. I wrote earlier about the seed corn the profession is eating: if nobody trains juniors, where do the seniors of ten years from now come from? And the one large employer still hiring juniors shows what the role has to become for that investment to pay off.
The short version: experience has to be built somewhere. And the work you used to build it with is exactly the work that is easiest to automate.
The senior is not safe
I regularly hear experienced developers say that AI mostly takes over the simple tasks.
Architecture, complex debugging, security and system design stay human work.
For now, a lot of that holds. The postings data says the same.
But it is dangerous to turn a temporary limitation into a permanent protection.
Today's models will look dated by 2028. And once agents can investigate a problem on their own, design a solution, run the tests and fix their own mistakes, the line moves again.
The senior postings point the same way. They ask for people who can work with AI: someone who directs agents, reviews their work and carries the responsibility. That is the role where you have to ask how many of them are needed.
Maybe there will always need to be a human who is accountable.
But how many?
One per five agents? One per twenty? One engineer overseeing an entire platform?
Nobody knows.
And anyone who claims with certainty that architects and seniors are safe is doing what the AI vendors do when they promise everything will run autonomously next year.
They are selling certainty they don't have.
More software does not automatically mean more jobs
The usual counter-argument is that cheaper software development creates more demand for software.
It might.
When computers got cheaper, we didn't use fewer of them. We found thousands of new uses. In June I called that Jevons paradox wearing a hoodie, and I still believe it holds for the demand for software.
But there is an assumption buried in that story: that extra demand for software also produces extra demand for human developers.
Those are two different things.
If AI drives production costs down far enough, the amount of software can explode while the number of people needed to make it falls.
More code. More applications. More systems.
Fewer developers.
Nothing in that combination contradicts itself. The Indeed numbers don't refute it either. A posting is not a job, a rebound from a trough is not long-term growth, and a market that mostly wants seniors is not growing in breadth.
My guess: 2030 gets uncomfortable
I don't expect every developer to be out of work by 2030.
What I do expect by then is a much clearer split between people who produce software and people who carry responsibility for what software does.
The first group will have a hard time.
The second will stay necessary for now, probably in smaller teams.
And somewhere between 2030 and 2035, the traditional picture of a software team might well disappear for good.
That is my estimate of where the technology and the economic incentives are likely to meet, and no reliable labour-market formula backs it up.
I could be five years off. Maybe ten.
Maybe demand for software turns out to be so large that we need more developers rather than fewer.
But I would not bet my career on that last scenario.
The uncomfortable answer
I work with AI agents every day. I see how impressive they are, and how many mistakes they still make.
I see code that appears in minutes and takes hours to review properly. I see duplication, missing context, and solutions that are correct locally but wrong at the level of the system.
Those are real limitations, and none of them protects your job.
If you want to be in the second group, make sure your work consists of what an agent cannot carry. Own a system all the way into production, and review what gets generated instead of passing it along. Be the person who can explain why something was built the way it was, because whoever merges the code answers for it, no matter who wrote it. And understand why judgement became the scarce part once speed got cheap.
For now, that is the side of the line where demand is growing.
The question is not when AI can do everything a developer can.
The question is when AI can do enough for a company to decide it can carry on with fewer developers.
And that moment may come a lot sooner.
We spent years building software to automate other people's work.
Now it's our turn.