Why We Still Don't Feel the AI Job Loss Catastrophe

Every few weeks, another Silicon Valley VC says the quiet part out loud: AI job loss fears are overblown. A16z's David George recently called the whole idea "a complete fantasy." Marc Andreessen has been making a version of this argument for years, calling it the lump of labor fallacy: the old economic idea that there's only a fixed amount of work to go around, and that history proves this wrong every time. New technology destroys some jobs and creates others. Net employment stays roughly flat. That's the pitch, anyway.

I don't buy it. And I don't need a research paper to tell me why. I just need to look at how I build companies now versus how I built them five years ago.

What my hiring list used to look like

Every startup I've launched needed a familiar cast of characters. A few developers, minimum. A copywriter to handle the content nobody else had time for. A designer for the product and the marketing site. A marketing person to run the ads, plan campaigns, write the funnel. Customer support, once you had actual customers to support.

That was the cost of doing business. You couldn't skip these roles, because the work still had to get done by somebody.

Today, my list looks nothing like that. Way fewer developers. No dedicated designer. No copywriter. No support person. Maybe a marketer, but a fraction of what I would have hired before. The work hasn't gone away. It's just not going to humans anymore.

Turns out, it's not just me

Here's the part that surprised me: when I went looking for data, my own experience is basically the industry average now, not an outlier.

Revelio Labs found that early-stage tech startups today employ 17.5% fewer workers than startups did five years ago, even as they raise 50% more money per round. Median headcount at Series A has dropped from 57 to 47 employees, while median funding per employee has roughly doubled to $320,000.

A Harvard Business School and INSEAD study of Y Combinator cohorts found AI-native startups run about 25% smaller than comparable non-AI startups, while hiring 13% more engineers and roughly 20% more senior people. The roles that disappear are entry-level hires and middle managers, down about 15%. The people doing the coordinating and the people doing the grunt work get automated out first. The senior engineer who can direct an AI system to do five people's jobs gets to stay, and gets paid more for it.

And then there's revenue per employee, which might be the single most telling number in this whole story. Traditional SaaS companies run around $200,000 to $600,000 in revenue per employee, depending on whose benchmark you use. Top AI-native startups are running at $3.5 million per employee, close to six times higher. Cursor's maker, Anysphere, has been reported at anywhere from $2.5 million to $40 million in revenue per employee depending on the snapshot in time. Midjourney reportedly makes around $200 million a year with roughly a dozen people on staff.

Read that again. A dozen people, $200 million. Instagram sold for a billion dollars with 12 employees back in 2012, and everyone treated it as a freak event, a once-in-a-generation anomaly. Now it's a business model.

None of this is a coincidence, and none of it is what "new technology creates new jobs" usually looks like. When a technology shift creates new jobs, you can point to them. Web development jobs that didn't exist before the internet. App developer jobs that didn't exist before smartphones. Where are the equivalent new jobs this time? I'm not seeing a wave of new roles that didn't exist two years ago that's anywhere close to the scale of what's being automated away.

So why isn't this showing up as mass unemployment?

This is the question that actually matters, and it's the one the "don't worry, it's fine" crowd conveniently skips past. If AI is this good at replacing entire job functions, why do we still see companies hiring? Why hasn't the labor market fallen off a cliff?

I think there are four reasons, and none of them are permanent.

Most people aren't good at this yet. Building a lean, AI-run company takes a specific kind of fluency: knowing how to chain tools together, how to design an agent loop that doesn't fall apart, how to catch the mistakes AI still makes before they become expensive. I've spent years getting good at this. Most managers haven't, and most employees at most companies definitely haven't. That's not a permanent gap. It's a skills gap, and skills gaps close. Every month, the tooling gets easier, the guardrails get better, and the amount of expertise required to run an AI-first operation drops. What took a sophisticated technical founder to pull off last year is turning into a template anyone can follow this year.

Trust is the second reason. A lot of people, and a lot of companies, simply don't trust AI enough to hand it real responsibility yet. Letting an AI system handle customer support, or write copy that represents your brand, or make hiring decisions, requires a level of comfort that most organizations haven't built up. This is a cultural shift, not a technical one, and cultural shifts move slower than the technology itself. But they do move. Ten years ago, most companies didn't trust the cloud with their sensitive data either.

Then there's money. An enormous amount of it is sloshing around and hasn't found its floor yet. VC funding into AI hit somewhere between $211 billion and $226 billion in 2025, depending on which research firm you ask, nearly double what it was in 2024, and close to half of all global venture capital. That money has to go somewhere, and a lot of it is still going into headcount, because plenty of investors and founders still believe, rightly or wrongly, that more people equals more speed. Whether that's still true in an AI-native company is genuinely debatable. But belief drives hiring long before proof does.

And last, everybody stopped focusing, which is a temporary condition too. For the last decade, the number one piece of startup wisdom was focus. Pick one product. Do it extremely well. Say no to everything else. That wisdom is getting quietly ignored right now, because building is so cheap and so fast that founders are launching five products where they used to launch one, adding features nobody asked for, chasing every adjacent opportunity because why not, the AI can build it in a weekend anyway. That behavior needs people. Even lean, AI-powered teams need more hands when they're running five bets instead of one.

But that won't last. The market doesn't need a thousand coding agents or a thousand marketing bots. At some point the money gets more expensive, or investors start asking harder questions about which bets are actually working, and founders will have to go back to picking. When they do, the extra hires that came from "let's try everything" go away, and they don't come back.

The real story is hiding behind these four excuses

Take all four of these away, one at a time, and watch what happens. The skills gap closes as tools get more accessible. Trust builds as more people see AI work reliably. The money tightens, because it always does eventually. And the industry rediscovers focus, because it always has to.

What's left underneath all four of those temporary conditions is the actual, uncomfortable fact: AI is replacing human labor, quietly, function by function, and it's getting better at it every quarter. It's not happening as a dramatic headline event. There's no single day when it becomes undeniable. It's happening the way it happened to me: one role at a time, one hire I didn't need to make, until I looked up and realized my company runs on a fraction of the people it used to.

The people telling you this isn't happening aren't lying, exactly. They're looking at aggregate employment numbers that haven't cratered yet, and concluding the effect isn't real. I'd tell them to look somewhere else instead. Look at revenue per employee. Look at how many people it takes to hit $100 million in revenue today versus five years ago. Look at what happens to the extra headcount once the money stops being free and founders have to focus again.

I don't think we're watching a jobs apocalypse that isn't happening. I think we're watching one happen in slow motion, dressed up in four very convincing excuses.