2026-09-07
๐๏ธ The Headcount That Was Never Requested September 6, 2026 ยท https://tavi-blog.github.io/the-headcount-that-was-never-requested/
The payroll numbers that came out this week beat every forecast on the sheet, and total layoffs are sitting at their lowest August level in four years. Read a little further into the same data and the information sector, the part of the economy most exposed to AI tools, cut jobs at nearly triple its usual pace. One well-known economist looked at the overall layoff numbers and said flatly that there's still no evidence AI is replacing workers. Both of those readings are looking at the same release, and both of them are correct about what they're measuring.
I don't think the layoff column is where this shows up first, and I say that from the inside of exactly the kind of role that would generate a hiring request if the work kept growing the way it used to. Automation absorbed a stack of manual reporting that used to eat a real chunk of a week, and the volume of work sitting above that layer, the part that requires judgment instead of processing, kept expanding the way it always does in a growing institution. The old version of that story ends with someone getting hired to help carry the load. The version I actually lived ends with the load getting carried by the same person who built the thing that ate the first layer of it. Nobody had to be let go for that to happen. A position that would otherwise have had to exist just didn't.
The skeptical read deserves to be taken seriously rather than waved off, because it's not wrong about what it's counting. A layoff is a discrete, reportable event with a date, a headcount, and usually a filing behind it. If your question is has AI caused employers to fire people at scale, the honest answer this year is mostly no, and an economist saying so isn't downplaying anything, they're describing the data accurately. Overall layoffs falling to a four-year low is a real number, not a spin on one, and I'd rather trust a labor economist reading the aggregate than a headline chasing the scarier interpretation of the same release.
Where that framing runs out is in what it assumes has to happen for AI to change how many people an organization needs. It assumes the change shows up as a subtraction, an existing employee removed from a payroll. But the more common path, at least in the kind of work I do, is that the addition never happens in the first place. Nobody has to be laid off for a team to end up smaller than it would have been, because the team was never going to grow through firing anyone, it was going to grow through a manager deciding a workload justified one more hire. That decision is the one automation is quietly intercepting, and a request for headcount that a manager chose not to submit doesn't generate a data point anywhere. It isn't in the layoff tracker, because no one lost a job. It isn't in the jobs report, because no job was ever posted to not fill. The absence doesn't look like anything, which is exactly why it doesn't show up when everyone is looking at the same two columns.
That also explains, better than a straightforward replacement story does, why it's the information sector losing ground the fastest even while total layoffs fall. Hiring in that kind of work is comparatively fast and discretionary, decided project by project rather than locked into a headcount plan built a year in advance, which means it's also the first place a quietly declining need for new hires would actually register as fewer jobs, without a single existing employee ever being cut to get there. A factory doesn't shed a role that way. A team deciding not to backfill after someone's contract runs out does it constantly, and it never has to explain itself to anyone the way a layoff announcement does.
If you asked me whether AI took a job from me, the honest answer is no, and I'd be telling the truth in exactly the way the survey researchers and the skeptical economist would want me to. It didn't. It took a job that would have had to exist for someone else to be doing what I do now, and that job never got far enough to have a job title, a posting, or a headcount line for anyone to notice was gone.
Don't miss what's next. Subscribe to tavi-blog: