AI software engineering jobs were supposed to disappear. The data just said otherwise.
For two years, software engineer job postings declined.
Every quarter. Consistently. January 2024 through late 2025.
The narrative made sense. AI writes code. So you need fewer people to write code. The headlines were confident. The career advisors were certain. The consensus was settled.
Then the postings reversed. They didn’t stabilize. They surged.
And Andrew Ng, co-founder of Google Brain, founder of Coursera, one of the people who built the infrastructure that trained the engineers who built the AI everyone is afraid of, called the advice that predicted this outcome “some of the worst career advice ever given.”
AI Software Engineering Jobs: What the Data Actually Shows
The data comes from Citadel Securities and Indeed, not a think tank with an agenda, not a tech company protecting its workforce. The firm that pioneered quantitative finance, tracking the actual job market.
Two years of consistent decline in software engineer job postings. Then a sharp reversal in early 2026. Then a surge.
The same week this data was published, new graduate software engineering postings exceeded 400,000 positions on a single job repository. GitHub Copilot had reached 20 million users. 84% of developers were actively using AI coding assistants in their daily work.
This is not the picture of a profession being eliminated. This is the picture of a profession being transformed, and expanded.
The AI that was supposed to eliminate the job created demand for a higher version of the job.
Why Andrew Ng Called It the Worst Career Advice Ever Given
Andrew Ng did not arrive at this conclusion from optimism. He arrived at it from history.
When computing moved from punch cards to keyboards, coding became easier. Every prediction said fewer programmers would be needed. The opposite happened, more people coded, not fewer, because the lower barrier to entry expanded the market for what software could do.
When the industry moved from assembly language to COBOL, articles published the exact same argument. COBOL is so easy that programmers are no longer needed. The opposite happened again.
His exact words: “As something becomes easier, more people should do it, not fewer. I think we’ll look back on that as some of the worst career advice ever given.”
The structural pattern is not a coincidence. It is how technology markets work. When the cost of producing something falls, the world does not order less of it. The world orders dramatically more — at a scale that was previously unaffordable. Lower cost expands the market. Expanded market increases demand. Increased demand requires more skilled people operating at a higher level.
The AI software engineering jobs surge is not an anomaly. It is the pattern, repeating exactly as it has every time before.
Who Is Actually at Risk, and Who Is Not
Ng is precise about this. He does not say AI poses no threat to software engineers. He says the threat is aimed at a specific group.
“It is true that a fresh college grad who is really on top of AI will outperform a full stack engineer with ten years of experience who is still doing things the way they were in 2022.”
Junior developer postings are down 40% versus pre-2022 levels. Computer science graduate unemployment sits at 6.1%, meaningfully above the national average of 4.3%. The people being hurt are not people who learned to code. They are people who learned to code but did not adapt, who are still writing boilerplate work that AI can generate in seconds.
And here is who is thriving.
“The best engineers I know are not fresh college grads. They are very experienced engineers who deeply understand architecture and the conceptual framework of how to think about computers, and additionally are on top of AI.”
Deep domain expertise plus AI fluency. That is the combination. A senior engineer who understands system architecture, security, performance, and how to direct AI to do the right thing in complex production environments is worth dramatically more today than three years ago, because they now have a 10x productivity multiplier in their hands.
One senior engineer described it precisely: “I’m not trying to out-code AI. I’m making myself essential by leading it with judgment.”
That sentence is the entire argument.
This Is Not a Story About Software Engineers
The software engineering data is the most visible proof point. But the pattern it reveals applies to every knowledge profession.
The people most at risk in the AI economy are not the people who kept learning and adapting. They are the people who stopped, because someone told them their skill was going to be automated, and they believed it.
Every profession has its version of the boilerplate work that AI can now do in seconds. The question for every professional is not whether AI can do that work. It is what they are building above it.
The head of technology at Jane Street, one of the most analytically rigorous trading firms on earth, was asked whether AI would replace his team. His answer: trading is AGI-complete. You cannot automate it without solving every other hard problem in the world. And then he said: “I have never been more desperate to hire more engineers and more traders than I am today.” Not less. More.
The AI software engineering jobs surge is not an isolated data point. It is part of a pattern that shows up everywhere when you look: Jensen Huang’s warning about the professionals who will be replaced, the Jevons paradox playing out in the Philippines and in every industry where AI has been deployed at scale, the Nature study showing that AI makes individual scientists more productive while shrinking the total volume of scientific ideas being explored.
The professionals who thrive are not the ones who competed with AI. They are the ones who built the five capabilities AI cannot replicate, and then picked up the tool.
The Five Capabilities AI Software Engineering Jobs Data Cannot Replace
After 27 years inside 34 Fortune 50 companies, I have identified five capabilities that remain structurally beyond what any AI system can replicate, regardless of how fast the engines get.
They are not soft skills. They are specific, buildable, measurable capabilities. And the AI software engineering jobs data confirms what each of them predicts.
IDEAS, the ability to see the connection no one has made yet, to ask the question no algorithm has been trained to ask. AI expands what is possible. It does not originate what is worth building.
SPEED, not the speed of execution, but the speed of recognition. The ability to read a moment, feel the window opening, and act before it closes. AI can execute faster than any human. It cannot feel the window closing.
TALENT, the deep conceptual fluency in a domain that Ng describes as the foundation the best engineers have. The judgment that comes from years of experience operating at the edge of a field. AI can replicate the output of talent. It cannot replicate the formation of it.
DISTINCTION, the ability to evaluate an answer and ask whether it is true, whether it is the right answer, whether the source is reliable. In a world where AI produces confident, authoritative output at scale, and is wrong with complete confidence nearly half the time on common questions, this capability is worth more than any tool that produces the output.
LEADERSHIP AT ALL LEVELS ‚Äî the accountability, judgment, and ownership that makes an organization’s AI investment worth what was spent on it. The organizations winning the AI era are not the ones spending the most. They are the ones whose leaders built human capability before deploying the tools.
The Andrew Ng argument, the Jane Street argument, the Citadel Securities data, they all point to the same place. The professionals who prepared are not being replaced. They are being promoted.
Where Do You Stand on the Five Ingredients?
The Kryptonite Scorecard is a 30-behavior diagnostic that tells you, specifically, behaviorally, and actionably, where you stand on each of the five ingredients AI cannot replicate.
The AI software engineering jobs data tells you the stakes. The Kryptonite Scorecard tells you where you stand.
It takes about 15 minutes. It gives you a clear picture of where you are prepared and where the displacement will find you first.
Take the Kryptonite Scorecard here.
Those prepared need not fear the forces at work.
Distinct or Extinct is available now on Amazon. Read more on AI jobs and the future of work: Jensen Huang’s AI jobs warning ‚ and what he left out | Microsoft Copilot adoption rate: why 96.7% are falling behind | What Is Your Number? The One Question That Tells You Where You Stand in the Age of AI