🗳️ The Survey Had Nowhere to Put My Answer August 31, 2026 · https://tavi-blog.github.io/the-survey-had-nowhere-to-put-my-answer/

A survey went around this week telling everyone to relax. Researchers asked more than a thousand US workers whether AI had cost them a job, gotten them a job, or gotten them promoted, and the overwhelming majority said no to all three. Something like nineteen out of twenty said AI hadn't taken their job. A much smaller share said it had created one for them or bumped them up a level. The headline write-ups landed on the same note: the doom and the hype were both getting ahead of the data, and for most people, work looks about the same as it did before any of this started.

I read through three separate write-ups of it looking for the box I would have checked, and it isn't there. My job title hasn't changed since before any of this, and by the survey's own definition I haven't lost anything, gained anything, or been promoted. I'd have to answer no three times, the same as someone whose work AI hasn't touched at all. That's not what happened. I built automations that took over a stack of manual reporting that used to eat a real chunk of a week, and the hours that freed up didn't sit empty and they didn't get me a new title. They went into the tier of work above the automation: judging what a flow produced instead of producing it by hand, tracking down why a report was quietly wrong for weeks before anyone caught it, scoping what an AI agent gets to touch in a regulated research process and what it doesn't. Nothing about that shows up as a loss, a gain, or a promotion. It shows up as the same job, doing something else.

There's a real defense of asking it the way they did, and I don't think it's a lazy one. Job loss and job creation are the two outcomes that actually show up in a labor statistic somewhere, a layoff notice, a new requisition, a title change with a pay band attached. If you're trying to settle a public argument that's mostly being fought with anecdotes and vibes, you want questions with an answer a payroll system could in principle confirm, not a question that asks someone to self-assess something as fuzzy as my job feels different now. Researchers picking measurable categories over squishy ones is usually the right call, and I'd rather see a survey do that than dress up a handful of dramatic quotes as data.

Where it stops being just a methodological trade-off is in what the finding gets used for once it's loose in the world. A result built to answer has AI been taking or creating jobs at scale is being read, headline after headline, as an answer to a much bigger question: is AI actually changing work. Those aren't the same claim, and the survey only speaks to the first one. A job can be entirely reshaped, hollowed out in one place and thickened in another, without a single person in that role losing it, gaining a new one, or getting a title bump, because none of that reshaping requires an employer to touch the org chart at all. My employer never had to file anything or announce anything for my actual daily work to look almost nothing like it did a few years back. The survey would count that as nothing happening. It's the opposite of nothing. It's just invisible to the three questions being asked.

I don't think the researchers were trying to make automation-as-transformation disappear. I think it disappeared because it's genuinely harder to survey. You can ask someone whether they got laid off in a sentence. Asking someone to accurately characterize how the content of their job shifted, without an employer-labeled event to anchor the answer to, is a much messier question to write and a much messier one to score, and I don't envy whoever has to turn that into a chart. But the absence of a good question for it isn't the same as the absence of the thing, and a result this clean, ninety-something percent reporting no change on every axis measured, is exactly the shape you'd expect if the thing actually happening to a lot of jobs right now doesn't live on any of those axes.

What I keep coming back to is that the loudest reassurance in the coverage, for most workers, nothing has changed, is true only if you accept the survey's definition of change. Mine changed. It just changed in a way that never had to ask permission from a title, a payroll line, or a headline writer looking for a clean number to lead with.