AI hasn't shown up in the jobs report yet, and that calm makes us uneasy

Hello fellow nerds,

Every morning our feeds tell us AI is about to change work forever. Then the monthly jobs report comes out and says, more or less, that not much happened. So, is AI taking jobs yet? Not in the official numbers so far. Two Google DeepMind researchers think that's because pay changes before headcounts do.

The latest report, from the US Bureau of Labor Statistics, counted 29,000 new jobs across the entire country in September. That's a slow month, and there's no wave of AI layoffs anywhere in it. Meanwhile the money behind AI keeps growing at a pace that's hard to picture. Alphabet just reported that Google Cloud revenue more than doubled from a year earlier. And this stuff is already part of ordinary workdays: Pew Research Center found that 38% of employed adults in the US use AI chatbots for tasks at work.

We'll be honest about how that mix feels to us. Part of us is relieved. If AI were wiping out jobs on a large scale, we'd expect to see it by now. Another part of us suspects we're looking at the wrong number.

The number that can't see it

This week, two people at Google DeepMind gave that second feeling some backing. On October 29, Alex Imas, Director of AGI Economics, and Shane Legg, co-founder and Chief AGI Scientist, published The Economics of Intelligence through the DeepMind Institute. The institute says its pieces are meant to start conversations and don't speak for Google, so we read it as two researchers making a case, with an economic model behind it.

Their starting point rang true right away. Most AI tests pit one model against one person, yet almost nobody works alone. Work happens in teams, and a good team beats its best member because people cover for each other's weak spots. Anyone who has run a small business knows the feeling. The person who's brilliant with customers is rarely the person you want doing the books, and the business works because both of them show up.

From there the authors split AI progress into two kinds. One is a single AI getting better at a skill, like writing or coding. They expect that to shift pay and prices gradually while companies stay roughly the same size. The other is many AI agents, meaning tools that carry out tasks on their own, getting good at working together. Past a point they call the supercooperative threshold, they expect "large, abrupt jumps" in market concentration, employment and wages.

The sentence we keep rereading is this one: "macro statistics of employment rates provide poor evidence that AI is not having an effect on the economy." In their model, the first stage looks calm from a distance because "all of the changes occur in wages." People whose skills AI can match would see their pay drift down toward what it costs to run the AI. Specialists, and the people who coordinate work, could earn more. Headcounts barely move, so the jobs report stays boring while individual paychecks change underneath it.

That was uncomfortable to read, because it means the calm number we felt relieved by can't tell us very much.

The calm number we felt relieved by can't tell us very much.

Where we land

We think the authors have the order of events right. Pay and job descriptions will change before headcounts do. Their sketch of today's market sounds familiar to us: leaner startups, flatter org charts, softer hiring for entry-level roles closest to what AI tests measure, and steady demand for experienced people who can direct complex work. They offer that as their reading, without measuring it. But if junior roles in your field have felt harder to land lately, this framework at least explains why that feeling and a flat jobs report can both be real.

Some changes are already in the official data, and they deserve care. BLS counts 129,000 fewer jobs in financial activities since May 2025, most of them in insurance. The bureau doesn't say why, and we won't pin it on AI. Those are real people and real paychecks. They're also a reminder that change usually lands on one industry or one kind of role long before it moves a national number.

Where we stay skeptical is the "all at once" part. It depends on AI agents getting much better at teamwork, and right now they struggle with it. The piece quotes OpenAI researcher Noam Brown on running thousands of agents at once: "it's actually very difficult to get these agents to coordinate in a productive way." The authors bet that won't last, and they admit nobody can say when, or whether, it changes, since today's benchmarks barely measure coordination. So we treat a sudden shift as a real possibility with no date attached. It's worth preparing for, calmly and without a deadline in our heads.

Our favorite part of the whole model is who turns out to matter most: the coordinator. On any solo test, a pure coordinator looks like the weakest person in the room, which will surprise no one who has ever tried to herd a group chat toward a decision. Inside a team, that person makes everyone else better, and in the authors' model that skill keeps its value.

What we're doing with this

That points to the most useful habit we can think of for anyone using AI, from people who opened ChatGPT for the first time last month to owners running a whole business on these tools. Get good at directing it. Give it clear jobs, check what comes back, and stitch the pieces together yourself. It also helps to write down the tasks you do in a normal week and mark the ones AI already handles well. That list will tell you more about your own job than any headline.

If you run a small team, we'd have the honest conversation early. Talk about which tasks AI is likely to take on, and plan training before anyone plans cuts. We'd want that kind of warning ourselves.

If you work for yourself, the model has some genuinely good news. The authors expect niche markets, where demand is small or custom, and what they call the "relational sector," where a human being is part of what people pay for, to hold up better against large AI firms. They point to healthcare, teaching, services and craft. If clients come to you for your judgment, your taste or the relationship, say so plainly on your site and in your pitch. That's the part of your work that's hardest to copy.

The October jobs report comes out on Friday, November 6. We'll read it, but we'll pay more attention to pay and entry-level hiring in the fields our readers work in, because that's where this framework says the story starts. We're uneasy and hopeful at the same time, and we suspect a lot of you are too. That seems like the right way to feel about something this big and this unfinished.

So here's the question we'll leave you with: in your own work, has AI already changed what a job pays or who gets hired, even if nobody has lost theirs yet?

— Artificially Nerdy