Last Updated on July 7, 2026 by Taya Ziv
For about two years, the story was settled. Software engineers were the dead men walking of the AI era. The logic felt airtight. AI writes code, code is what engineers make, so engineers are the first job the machine eats. Dario Amodei said the quiet part into a microphone last year, warning that AI could wipe out half of all entry-level white-collar work and push unemployment toward 20%. Every founder I talked to had quietly filed engineers under “roles I can stop hiring soon.” May 2026 seemed to prove it, with the worst single month of tech layoffs in years, and AI cited as the reason more than anything else.
So here’s the part nobody put on a slide. The data just came in, and engineering wasn’t the first job to fall. It was the last one standing.
What the numbers actually say
SignalFire is a venture firm that does something most VCs don’t bother with. They track the real career moves of millions of people across more than 80 million companies, and they read hiring data instead of layoff announcements, because layoffs are noisy and people sit on the news for months before they update a profile. Hiring is the cleaner signal. It shows what companies are actually doing with their next dollar, not what they’re telling the press.
Their State of Talent Report for 2026 lands on a number that should stop you cold. Across the twelve companies they call the Tech Majors, the Googles and Metas and Nvidias of the world, engineers made up 55% of all new hires in 2025. Back in 2019, before any of this, that figure was 46%. So in the exact window when AI was supposed to gut engineering, engineering’s share of hiring went up, not down. Total tech hiring across those giants dropped about 25% compared to 2019. Engineering roles dropped only 11%. Everything else got cut harder than the job everyone said would go first.
And it’s not just the giants protecting their crown jewels. Early-stage startups, the companies with no budget to waste and every reason to skip a hire if AI could cover it, actually brought on 7% more engineers in 2025 than they did in 2019. These are the most cost-paranoid organizations on earth. If a robot could really do the job, they’d have been the first to notice and the first to act. They went the other way.
Why this matters if you’re the one signing the offers
This isn’t trivia. If you’re a founder right now, you are making one of the most expensive decisions of your year, and you might be making it on a headline. The question is simple and it costs real money to get wrong. Do I hire an engineer, or do I bet that AI lets me skip one?
A lot of founders have been quietly choosing “skip.” They read the same doom I did. They saw the same prediction that had AI agents eating software jobs by the dozen, and they decided the smart, lean, modern move was to run on tooling and a single technical co-founder for as long as humanly possible. And the cleanest hiring data we have is sitting here telling them that the companies who win are doing the opposite. They’re hiring the engineers. They’re just getting a lot more out of each one.
We’ve actually seen this exact shape before, and recently. We already watched the company everyone called a job-killer turn around and hire four thousand people, which made no sense if you believed the machine simply subtracts humans. It makes perfect sense once you understand what’s really going on.
The boring economics nobody wanted to mention
There’s a 160-year-old idea hiding under all of this, and it has a name. The Jevons paradox. In the 1860s an economist noticed that when steam engines got more efficient and coal went further, England didn’t burn less coal. It burned way more. Cheaper, more useful coal meant everyone found new reasons to use it, and total demand exploded. Efficiency didn’t shrink the market. It grew it.
That’s engineering in 2026. AI didn’t make writing code worthless. It made writing code cheap, and cheap means people suddenly want a lot more of it. Every feature you shelved because it wasn’t worth a month of dev time is now worth building. Every internal tool, every integration, every “wouldn’t it be nice if” is now on the table because the cost of trying dropped through the floor. Jensen Huang put it plainly when someone asked him if AI was going to destroy software jobs. He said the opposite was happening, that his engineers were busier than ever, because the agents write code instantly and then turn around and demand the next idea, and the next, faster than humans can feed them.
So the value didn’t disappear. It moved. It slid off the thing AI is good at, which is producing lines of code, and onto the thing AI is still bad at, which is deciding what’s worth building and noticing when the machine is confidently wrong. The grunt work got automated. The judgment got more valuable. And judgment is still wearing a human face.
The trap hiding inside the good news
Now let me ruin the happy story a little, because if you stop reading here you’ll draw the wrong lesson.
The resilience is real, but it’s lopsided. The same data shows companies are compressing into a senior-heavy engineering core. They’re keeping and hiring the experienced people, because an experienced engineer plus AI is genuinely a small army. And to pay for that, a lot of them are quietly shutting down the thing that felt most expendable, the new-grad pipeline. Why train a junior for two years when a senior with AI covers the output today? It pencils out beautifully on this quarter’s spreadsheet.
It’s also how you build a slow-motion disaster. If the whole industry stops hiring juniors at once, there’s no one in the pipe to become the seniors everybody’s fighting over in five years. You’re eating your seed corn and calling it efficiency. This is the same uncomfortable thread I keep pulling at, that the easy AI math almost always punishes the future to flatter the present.
My take, and where I might be wrong
Here’s what I actually believe. The engineering apocalypse was always more anxiety than analysis. It was a story that felt true because the technology was genuinely shocking, and we mistook “this is powerful” for “this replaces people.” Those are not the same sentence. Powerful tools usually make skilled people more valuable, not less, and they have for most of industrial history. The AI-jobs panic forgot a century of economics because the demo was too impressive to argue with.
For founders the move follows from that. Don’t run your company on the doom headline. Hire the engineer who turns AI into ten engineers, and stop indulging the fantasy that an AI agent can just be your startup’s first employee while no human is steering it. The agent is a power tool. Power tools don’t decide what to build, and they certainly don’t notice when they’ve spent three days building the wrong thing beautifully.
Now let me argue with myself, because I can feel the version of this where I’m wrong. The data is about companies that are scaling, that already have product and customers and a reason to ship more. It is not about you in week one with a Figma file and a dream. At that stage, adding engineering headcount before you’ve proven anyone wants the thing is the oldest mistake in the book, and AI tooling genuinely does let a single technical founder go further than ever before. So “engineers are resilient” is not a license to hire a team before you have demand. The honest version is narrower. Once you have something real to build, the instinct to skip engineers because AI will handle it is backwards. Before that, restraint still wins. And I’ll admit the senior-heavy trend could mean the door is harder to walk through for exactly the junior engineers a small startup can actually afford, which complicates the clean “just hire engineers” advice more than I’d like.
What to actually do about it
Three things, and none of them is “wait and see how the AI thing shakes out.”
First, stop treating engineering as the cut you make to look lean. The data says it’s the function compounding fastest in value, so cutting it first is cutting muscle and calling it fat. If you’re trimming, trim somewhere the machine actually replaced, not the place it made stronger.
Second, hire for judgment, not typing speed. The engineer worth paying for in 2026 is the one who’s good at deciding what to build and catching the AI when it’s confidently wrong, not the one who can produce the most code by hand. That changes who you interview for and what you test. Ask them to direct AI and spot its mistakes, not to whiteboard a sorting algorithm from memory.
Third, if you can possibly afford one junior, take the bet everyone else is skipping. The whole industry is hoarding seniors and starving the pipeline, which means good early-career engineers are about to be undervalued and available, and you can grow your own seniors while your competitors fight over the same expensive ten people. Scarcity at the top usually means a bargain in the middle, if you’re brave enough to invest a year ahead of everyone else.
The job that wouldn’t die
I keep coming back to how confident we all were. Engineers were finished, it was obvious, the machine writes the code now. And then the only people reading actual cap tables and hiring logs instead of headlines looked up and found engineering was the single hardest job to kill in the entire tech economy. The thing that was supposed to go first didn’t go at all. It just got a power tool and a longer to-do list.
The lesson isn’t really about engineers. It’s about how easily a scary, impressive technology talks us out of boring, durable economics. Cheaper doesn’t mean gone. More efficient doesn’t mean fewer. Most of the time it means more, and the people who can wield the new thing well are worth more than they were yesterday, not less. So before you skip the hire because AI has it covered, look at what the companies with the best information are actually doing with their money. They’re hiring the humans. They’re just handing each one a much bigger lever.
