For two years I told people the layoff email was the signal. I was wrong, and I was wrong by roughly two quarters.
The thing that actually predicts AI job displacement in customer-facing work is almost never the announcement. It is a title. A new VP of Customer Experience Transformation. A Chief AI Officer with a dotted line to the COO. A support organization that stops reporting to Operations and starts reporting to Product. By the time the all-hands happens, the decision is eight or nine months old and has a slide deck with a number on it.
No affiliate links in this one. There is nothing here for me to sell you, which is relevant, because a lot of what gets written on this subject is written by people who need you to buy a certificate by Friday.
The question you have actually been asking
Not "will AI take my job." That question is too large to answer and too vague to do anything with. The question you have been asking, at 11pm, in the version of your head that isn't performing optimism for anyone, is narrower than that:
Is AI actually replacing my work, or is "AI" the word my company is going to use when it does the thing it already wanted to do?
The honest answer, in one sentence: it is usually both at once, and the ratio between them matters far more than the label — automation is genuinely absorbing a narrow band of high-volume, text-shaped tasks, while the language of AI is doing much heavier lifting as cover for cost decisions that were on the table before a single model got deployed.
That is not a dodge, because the ratio is knowable. Not perfectly. But you can read most of it off the org chart, months before anyone sends you a calendar invite with no agenda.
What leadership changes tell you that headcount numbers don't
Headcount is a lagging indicator. It reports what already happened. Reporting lines are a leading one, because budget follows the box on the org chart, and a newly created box needs to justify itself inside about four quarters.
This is the mechanism nobody explains to you. When a company hires an executive whose entire mandate is "AI transformation," that person does not arrive with a neutral question about whether AI helps. They arrive with a thesis they were hired for. Their first review cycle requires a number. The fastest available number in a support organization is not revenue and is not customer satisfaction — it is cost per contact, and the cheapest way to move cost per contact is fewer people answering contacts.
The public record on this is less mysterious than it looks. In 2023 IBM's chief executive said publicly the company would pause or slow hiring for back-office roles it expected AI to absorb, naming a figure in the thousands. In 2024 Klarna's chief executive said the company's AI assistant was handling the work of roughly 700 agents; by 2025 he was saying publicly that quality had suffered and the company was hiring human agents again. In 2025 Salesforce's chief executive said support headcount had gone from around 9,000 to around 5,000, crediting AI agents. That same year, Duolingo's "AI-first" internal memo leaked, the backlash was loud enough that leadership walked the framing back, and the underlying strategy stayed roughly where it was.
Four companies, four different outcomes, one consistent pattern: the leadership statement came first, the headcount move came second, and the correction — where there was one — came third and quietly.
How I'd actually read the signal
Four criteria. I use them in this order, and each one has a limit I'll name, because a signal you can't falsify isn't a signal.
Where your function reports now
If customer support reports to Operations, you are a cost center being managed. If it moves under Product or Engineering, you have been reclassified as a data source and an interface problem. That reclassification is the single strongest early indicator I've found, because it changes who writes your roadmap and what they think you are for.
The limit: plenty of companies move support under Product for genuinely good reasons and get better tooling out of it. The reporting change tells you the frame has shifted. It does not tell you the direction.
Where the new leader came from
An executive promoted out of operations knows what your queue looks like at 4pm on a Monday. An executive hired in from a platform company or a consultancy knows what your queue looks like in a dashboard. That difference shows up in every decision they make for the next two years.
The limit: this reads as a character judgment and it isn't one. Some outside hires are the only people with enough standing to protect a team during a budget fight. You are reading incentives, not virtue.
Whether your work is already measured in tickets
This is the uncomfortable one. Roles that were already quantified into per-unit metrics — handle time, contacts per hour, first-contact resolution — were pre-formatted for automation years ago. The measurement came first. Automation follows measurement the way water follows a channel someone else dug.
The limit: the tasks that are easiest to count are frequently not the tasks that keep customers. Companies discover this after the cut, and the rehiring is always smaller and worse-paid than what it replaced.
Whether the hiring freeze is quieter than the layoff
The cut gets a press cycle. The freeze gets nothing. But a company that stops backfilling attrition in your function is telling you more than a company that cuts ten percent and says the word "AI" four times in a memo. Watch the requisitions, not the announcements. If your team has lost three people in a year and hired none, the decision has been made and nobody had to write anything down.
The limit: freezes reverse. Backfills resume when a quarter goes well. This signal is noisy over short windows and reliable over eighteen months.
Where the answer is genuinely "it depends"
I'm not going to pretend the picture resolves cleanly, because it doesn't.
It depends heavily on whether your role has ever been offshored. Work that survived an offshoring wave survived because something about it resisted decomposition — regulatory exposure, relationship continuity, judgment under ambiguity. That same property tends to resist automation, at least for now. Work that was offshored and then brought back for quality reasons is in the most exposed position of all, because the cost case has already been made once internally and the file is still open.
It depends on whether your employer sells software. Companies that sell AI tools have a commercial reason to describe their own workforce reductions as AI success stories, whether or not that's what happened. Companies that don't sell AI have no such incentive and tend to describe the same cuts as "restructuring." Same action, different press release, and the press release is what most reporting is built from.
And it depends on something you cannot see from your desk: whether the cut was already scheduled. In a real number of cases, the model was the justification, not the cause. That distinction changes nothing about your paycheck and everything about how much you should trust the stated reason when you're deciding what to do next.
Who should be moving now, and who shouldn't
Move now if your function was reclassified in the last year, your leadership changed, and backfills stopped. That combination has a high hit rate and a long enough lead time to be useful. Move meaning: build the internal record of what you do that isn't a ticket, and start looking while you still have leverage and a badge.
Don't move on a single signal. A new Chief AI Officer alone, with no reporting change and normal backfilling, is a company buying a hedge, not executing a plan. Panic-quitting a stable role into this market on one data point is a worse outcome than the thing you're avoiding.
The uncertainty is not your failure of analysis. It is the actual condition, and anyone selling you certainty about it is selling something.
Watch the org chart, not the memo. The memo is written after the decision.
