In the eleven months I was out of work I sent 288 applications and got seven phone screens. Somewhere around application 150 I started keeping a second spreadsheet. Not of jobs — of tech layoffs. One row per announcement at a company where I had something pending, with a column for the date, a column for how many people, and a column I labeled "reason given."
That last column is the one that ruined me. I can't read a corporate announcement like a normal person anymore. I read it like a diff.
Here's what the column looked like after nine months. Same-sized cuts, same quarter, four different stories: one company had over-hired during the pandemic. One was flattening its org to move faster. One was responding to macro conditions. One had adopted AI. Two of the four had beaten their own earnings guidance that same month.
The verdict, up front: when a company names AI as the reason for its job cuts, the AI is usually a description of the future it wants priced into the stock, not an account of why the payroll shrank last Tuesday — the reason is nearly always cost, and cost has been the reason since 1993. That does not mean AI isn't changing your work. It means the press release is not evidence about it.
No affiliate links in this one. There is nothing here to buy. I am not selling you a course on how to read layoff announcements, which would be an unhinged thing to sell to someone who is unemployed.
Where the word came from
Every layoff era has a load-bearing word, and the word arrives before the cuts do.
In 1990 a computer scientist named Michael Hammer published an article in Harvard Business Review titled "Reengineering Work: Don't Automate, Obliterate." Three years later he and James Champy turned it into a book, Reengineering the Corporation, and it sold in numbers business books almost never reach. The argument was about process, not people. You weren't supposed to pave the cow paths with computers. You were supposed to redesign the work from scratch.
What happened instead is that "reengineering" became the word you said when you meant "we are cutting thirty percent of the staff." The doctrine was about process design. The deployment was severance. By the mid-nineties Hammer was telling reporters he had underestimated the human side of it — that he had been reflecting his engineering background and hadn't been appreciative enough of the people inside the diagrams. That admission arrived after the word had already been used to justify hundreds of thousands of cuts. The correction never travels as far as the claim.
Then 1996. AT&T announced 40,000 cuts and Newsweek ran a cover of executive headshots under the words "Corporate Killers." The most fluent spokesman for the entire doctrine was Al Dunlap, who cut his way through Scott Paper, went to Sunbeam, and wrote a book arguing shareholders were the only constituency that counted. He was later charged by the SEC over Sunbeam's accounting and settled without admitting wrongdoing, agreeing never again to serve as an officer or director of a public company. The most articulate voice for cutting-as-strategy turned out to be a man inflating the numbers.
After that the words rotate on a schedule. 2001: right-sizing, realignment. 2008: the macro environment. 2020: staffing for the world we expect. 2022: we over-hired. 2023: the year of efficiency. Now: AI.
You've lived through at least two of these. You know the rhythm even if nobody's laid it out for you: the word shows up in one letter, gets repeated on three earnings calls, and within a quarter every company in the sector is using it in the same paragraph position. The word is not a discovery. It's a genre convention.
The part where the source turns out to be thinner than the belief
Here's what I didn't expect to find when I started the spreadsheet.
Nearly everything you and I know about the current wave of tech layoffs traces back to companies describing themselves.
The number you've seen cited most — the running total of tech workers cut since 2022 — comes in large part from layoffs.fyi, a tracker one person, Roger Lee, began assembling in March 2020 out of news reports. It's genuinely useful and it's honest about what it is. It is also a spreadsheet of press coverage, and press coverage of a layoff is mostly a rewrite of the company's own announcement.
The other number comes from Challenger, Gray & Christmas, which publishes a monthly count of announced job cuts broken out by stated reason. Read that phrase again. Stated reason. There's a category for AI, a category for cost-cutting, a category for market conditions, and the thing that determines which bucket a cut lands in is what the employer said in its own announcement. The employer picks the bucket. Then the number gets aggregated, then reported, then cited as though somebody audited it.
State WARN filings are the closest thing to a primary document. They'll give you job titles and site addresses, which is more than the blog post does. They will not tell you why.
So the belief is "AI is eliminating tech jobs at scale," and the source, at the bottom of the stack, is companies saying so, in documents written to be read by people who own the stock.
And when someone does attach a number to the AI claim, watch what happens to the number. In early 2024 Klarna's chief executive said the company's AI assistant was doing the work of 700 full-time agents. That figure went everywhere. It's still circulating. By 2025 the same executive was saying publicly that they had cut too far, that quality had suffered, and that human agents were coming back. The retraction got a fraction of the pickup the 700 did. IBM's chief executive said in 2023 that the company would pause hiring for several thousand back-office roles it expected AI to absorb; a couple of years later the account had become that a few hundred HR roles were automated while the company hired more engineers and salespeople. Both are real events. Neither supports the belief in the shape the belief has been used.
I'm not telling you the technology is fake. I'm telling you that the evidence for the specific claim — this layoff, these people, because of AI — is, as of writing, a set of sentences produced by people with an obvious incentive to produce them.
The four reasons, graded
If you're deciding where to apply, whether to take an offer, or whether to stay, you need to grade these rather than dismiss them. Every explanation for tech layoffs has a legitimate version and a decorative one. Each of the four is true sometimes. Each has a tell.
"AI is making us more efficient"
What's real: in support, QA, content operations, first-line triage, and first drafts of a lot of code, output per person has genuinely moved. If your work is producing a first version of something a human then edits, that has changed and you already know it.
The honest negative: this is the only reason on the list that raises the share price, and the only one that can't be checked from outside. No company publishes a per-role attribution. "AI-enabled efficiency" in an announcement is a forward-looking statement wearing the costume of a historical one. Watch for the hedge that usually rides along with it — the line clarifying that the departing people are not being replaced by AI. That sentence is doing two jobs at once. It keeps AI in the paragraph for investors and out of the paragraph for employment lawyers.
"We over-hired"
What's real: often literally true and, unusually, checkable. Headcount is disclosed in annual filings. If a company tripled staff between 2020 and 2022 against a demand curve that then flattened, the correction isn't a mystery. The clearest version was Mark Zuckerberg's November 2022 letter, which said in plain words that he had expected the pandemic shift to online to be permanent, that he'd been wrong, and that it was his call. Whatever else you think about that company, the sentence was falsifiable and it named a person. Salesforce cut about 8,000 people two months later, roughly a tenth of the company. Amazon's announcements across that winter added up to about 27,000. The over-hiring account fit most of them.
The honest negative: it's a story about a mistake that has now been corrected, which is precisely the impression a company wants to leave and precisely the impression that keeps turning out to be premature. It also assigns blame upward, which is why it reads as candid and why it gets used once and quietly retired. Nobody says "we over-hired" three years running.
"Flattening, agility, focus"
What's real: management layers accumulate, and a company with eight levels between an engineer and a decision is genuinely slow. Capital One cut roughly 1,100 technology roles in January 2023 and said its agile transformation had reached a point where an entire job family was no longer needed. That is a coherent thing to have concluded, if a brutal one.
The honest negative: this is the emptiest of the four. It's the phrasing you reach for when the answer is cost and you'd rather not say cost. In my spreadsheet it also carried the worst repeat rate — the companies that said "flattening" were the ones most likely to reappear in the column six or nine months later.
"Macro conditions"
What's real: rates, ad spend, enterprise budgets, and demand cycles are external and verifiable. If every company in the same segment is contracting at once, that's a live explanation rather than a shrug.
The honest negative: it outsources causation completely, and it sits badly next to a quarter the company beat. Wayfair cut about 1,650 people in January 2024, some months after a memo from its chief executive about working longer hours and blending work and life. When the macro story and the culture story are that close together on the calendar, one of them is doing different work than it claims to.
The fifth reason, which is not in the announcement
The cut lands in Boston, Austin, and Seattle. The reqs open in Hyderabad, Warsaw, Guadalajara, and Costa Rica. Nobody says this on the earnings call because there's no version of it that reads well in a headline, but it shows up on the careers page within a quarter if you're watching for it. Of everything in this piece, this is the one most likely to be the actual reason your application went nowhere.
| Reason given | Usually true when | The tell | What it means for your application |
|---|---|---|---|
| AI efficiency | Support, QA, content ops, first-draft work | No number attached; paired with "these roles aren't being replaced by AI" | Senior reqs continue; entry-level postings quietly stop appearing |
| We over-hired | Headcount tripled 2020–2022, demand flattened | Used once, then never again by that company | Few backfills for two or three quarters, new headcount rarer |
| Flattening / agility | Eight layers between you and a decision | Vaguest wording, highest repeat rate | Expect another round; ask where the role reports |
| Macro conditions | The whole segment contracts at once | Company beat its own guidance the same month | Slow loops, frozen reqs, offers with short fuses |
| Location arbitrage (unstated) | Cuts concentrated in high-cost metros | Same job family reposted abroad or a band lower in 30–90 days | Your req may be the cheaper version of someone's old job |
How I'd actually decide
Six checks. They take about twenty minutes per company. I ran them before every interview I got.
Check the calendar first. Pull the earnings date and the announcement date. Cuts announced within a few days of earnings are about a number in a spreadsheet, whatever the paragraph says. Cuts announced in the dead middle of a quarter are more likely to be structural.
Read the 8-K, not the blog post. US public companies file one for a material restructuring, and it names the severance and charge figures. That's the number the company is legally careful about. The blog post is the number the company is rhetorically careful about.
Compare total headcount year over year. It's in the annual report. A ten percent cut alongside flat total headcount is a replacement, not a reduction. That's the most useful number in this entire process and almost nobody looks it up.
Open the careers page thirty days later. Filter to the same job family. If the roles are back at a lower band, or in a different country, you now know something the announcement declined to tell you.
Ask whether the AI claim has a number attached. "AI is letting us do more with less" is a mood. "We cut ticket handle time by X and reduced the support org by Y" is a claim. Claims can be wrong. Moods can't be anything.
Count how many times the same company has appeared. Three announcements in eighteen months means the plan isn't working, regardless of which word got used each time.
What to do with this in a phone screen
Two questions. Ask the recruiter, not the hiring manager, and ask them flatly, without apologizing for asking.
"Is this role a backfill or new headcount?"
"Was this team affected by the reduction in [month]?"
Recruiters answer these. They answer because the questions are ordinary, and a recruiter who dodges both has told you what you needed anyway. While you're there, look at the posting date and whether the req has been refreshed; a role that's been open and re-posted for five months is either a pipeline posting or a team that can't get approval to close.
None of this gets you hired. What it does is stop you spending your two good hours a day on companies performing a hiring process rather than running one. At 288 applications, that is not a small thing.
Who this is for, and who it isn't
This is for you if you're mid-career, you've been through more than one round of tech layoffs, and the announcements have started to read like a language designed to be nodded at rather than understood. It's for the person weighing an offer from a company that cut last quarter and trying to work out whether the seat is real.
It isn't for you if you want a forecast. I can't tell you whether your role exists in five years, and anybody claiming the announcements answer that is reading marketing as research. It's also not for you if you want to walk away concluding the technology is nothing. Everything here is about the gap between a claim and its evidence, not about whether the tools work. Those are two different arguments, and the second one won't be settled by press releases either.
What I couldn't answer
Whether this time is different.
That's the real question and I don't have it. It's possible the words have rotated for thirty-five years for the same cost reasons and this is one more rotation with better graphics. It's also possible the word finally caught up to reality, and the first genuinely automation-driven contraction in white-collar work is happening right now behind sentences that used to mean nothing. From inside a press release those two look identical, because the press release was never built to tell them apart.
So here's where I'd look instead. Job postings rather than announcements — specifically the ratio of junior to senior openings at one company over eighteen months, because if AI is eating anything first it's the bottom rung, and a careers page admits that before an executive will. State WARN databases, which list job titles instead of narratives. The federal occupational employment surveys, which are slow, dull, a year behind, and compiled by people with no share price to protect. And the reopened reqs — the quiet ones, in a different city, four months after the cut.
The announcement is written for people who own the company. The job posting is written for people who have to work there. Read the second one.
