The Approval-Step Layoff: Why 316,000 AI Job Cuts Landed On One Exact Kind Of Job
Every headline about artificial intelligence and jobs is getting the story half right and the important half wrong.
Here is the number everybody quotes. Since 2023, more than 316,000 jobs have been cut by companies that named artificial intelligence as the reason. In 2026 alone, tracking through mid July counted 267 separate layoff events touching roughly 185,894 workers, and about 56 percent of those events cited AI or automation as a driving force.
That is a real number and it belongs to real people. But it is not the story.
The Story Is Not How Many. It Is Which.
Look at the actual list of what got cut.
Data entry. Bookkeeping. Payroll clerks. Administrative assistants. Customer service representatives. Junior analysts. Entry level software engineers.
Read it twice, because there is a pattern sitting in plain sight and almost nobody has named it.
Every single one of those is a job where a human being was the approval step.
Somebody typed it in. Somebody checked it against the source. Somebody verified the number before it went out the door. Somebody signed off. In each of those roles, the person was the checkpoint standing between an input and a consequence.
That is not a coincidence, and it is not a story about coding. It is a story about checkpoints.
When a company automates the approval step, it does not eliminate the work. It eliminates the person whose entire job was to be the approval step.
I call it the approval-step layoff. Once you have the phrase, you will not be able to stop seeing it.
The Receipts
This is not a theory built out of vibes. The specifics line up almost too neatly.
IBM's chief executive confirmed that AI agents replaced several hundred employees in the company's human resources department. Their internal system now handles roughly 94 percent of routine HR tasks. Routine HR is, almost by definition, a series of checkpoints. A request comes in, somebody confirms it against policy, somebody approves it, somebody records it.
Klarna reduced customer service headcount by roughly 700 agents after deploying AI that now handles around 70 percent of customer interactions. Customer service is a checkpoint too. A person arrives with a problem, somebody verifies their account, somebody decides whether the request qualifies, somebody authorizes the resolution.
Major outsourcing firms have cut data entry headcount by 30 to 40 percent since 2024, as document processing tools made manual entry obsolete. Data entry is the purest checkpoint that exists. A human being looks at one representation of information and confirms it into another.
Amazon displaced more than 16,000 warehouse and logistics roles with robotics. SAP moved 8,000 roles inside a transformation. BT Group automated 10,000 customer service positions.
Different industries. Different technologies. Same underlying job.
Why This Distinction Actually Matters To You
If you believe the story is "AI takes jobs," you end up in one of two useless places. Either you panic about a collapse that the data does not support, or you dismiss the whole thing because your job still exists.
If you understand the story is "AI absorbs approval steps," you can do something with it. You can look at your own week and ask a specific, answerable question.
How much of what I get paid for is checking, and how much is deciding?
Those sound similar. They are not remotely the same thing.
Checking is verifying that something conforms to a known standard. Deciding is choosing what to do when the standard does not cover the situation, and then being accountable for that choice.
A machine is extraordinary at the first one. It is structurally incapable of the second, because accountability cannot be assigned to software. Somebody has to be able to lose something when it goes wrong.
The Cruelty Hiding In The List
There is one entry on that list that deserves its own paragraph, and it is the one that keeps me up.
Entry level software engineer.
That is the rung. That is the exact rung people climbed to get from a paycheck to a career in this economy. The same is true of junior analyst, of assistant, of clerk.
Here is the trap. Those roles were never really about the output. Nobody needed a 23 year old to produce marginally acceptable code. What those roles produced was judgment, slowly, in a human being. You did the junior work, you made mistakes in a low-stakes environment, you learned what good looks like by producing bad and being corrected.
We are removing the training ground and keeping the requirement.
A kid graduating this year is being asked to arrive senior, in a field where the junior work was the mechanism that made people senior. We are pulling out the bottom of the ladder and telling 22 year olds to jump higher.
That is not a technology problem. That is a choice companies are making, and it will show up as a talent shortage in about 6 years, right on schedule, and everyone will act surprised.
The Oversight Paradox
The people running these cuts have already tripped over the flaw, and some of them know it.
Companies are eliminating the workers who hold the institutional knowledge, while simultaneously lacking a trained workforce capable of directing, validating, and quality-controlling the AI systems meant to replace them.
Go back to that IBM figure. The system handles 94 percent of routine tasks. Sit with the remaining 6 percent for a second.
That 6 percent is not the leftovers. That is where every hard case lives. The exception. The dispute. The situation the policy did not anticipate. The one where getting it wrong costs the company a lawsuit.
And the people who knew how to handle that 6 percent were disproportionately in the group that got cut, because on a spreadsheet their day looked like routine processing.
This is why several companies have quietly started rehiring into roles they eliminated 18 months earlier, usually with a new title and a higher salary. The work did not disappear. The ability to do it did.
What This Means If You Run A Business
I deploy this technology for a living, in real estate and in local business operations, with my own money on the line. So let me be practical instead of philosophical.
The opportunity is real and I am not going to talk you out of it. Automating a genuine checkpoint is often the highest return move available to a small operation. If a person on your team spends 6 hours a week confirming that one system matches another system, that is work a machine should be doing, and freeing that person up is good for them and good for you.
But run this test before you switch anything on.
Ask what happens when it is wrong. Not whether it will be wrong. It will be. Ask what the wrong answer costs, who finds out, and how fast. If a mistake produces a slightly awkward email, automate freely. If a mistake produces a compliance problem, a lost client, or a number on a contract, keep a human on the checkpoint and use the machine to prepare the work instead of finalizing it.
Ask who is accountable. Not who is responsible in an org chart sense. Who takes the hit. If the answer is "the system," you have not removed the approval step. You have hidden it, and it will come back at the worst possible moment.
Ask whether the checkpoint was teaching somebody something. This is the one almost nobody asks. If the role you are about to automate is how your next senior person was going to get built, you are not saving money. You are borrowing it from yourself at a terrible rate.
The Part Nobody Wants To Say Out Loud
None of this is being decided in public.
There is no vote on whether the approval step goes away. It goes away as a default setting in a product update. It goes away as a line item in a reorganization. It goes away because a competitor did it first and the pace of a race gets set by competitors rather than by readiness.
Every trade in America already knows that feeling. It is the safety meeting that gets shortened because the job is behind schedule.
And the pressure only goes one direction, because the money only goes one direction. When a company takes on obligations measured in tens of billions of dollars for compute alone, it acquires a personality whether it wants one or not. It has to grow, it has to ship, it has to justify the burn on a schedule. A company carrying that weight cannot afford to be the careful one. The math will not permit it.
So the careful part has to come from somewhere else. It has to come from the people actually deploying this stuff on the ground, in small businesses, in offices, in one-person shops.
That is not a consolation prize. That is genuinely where the decision lives now.
What To Actually Do This Week
Not a five year plan. One thing.
Find a single place in your own work where you are the approval step, and decide on purpose whether that should stay true.
Some of them should. The ones with liability attached, the ones where a client is trusting your name, the ones where you would not be able to explain to somebody why a machine made that call. Keep your hands on those and say so out loud, because that is worth something in a market where everybody else is quietly handing them over.
Some of them should not. The ones where you are confirming that one number matches another number, where your judgment is not actually being exercised, where you are functioning as expensive human glue between two systems. Let those go and use the time on work that only a person can be accountable for.
The distinction between those two categories is the most valuable thing you can figure out about your own job this year.
Because the machines are extraordinary at checking, they are getting cheaper at it every quarter, and the deciding is the part that is still yours.
It is only still yours for as long as you keep noticing when somebody offers, very politely, to take it.
Common questions
How many jobs have actually been lost to AI?
More than 316,000 since 2023 have been cut by companies naming artificial intelligence as the reason. In 2026 alone, tracking through mid July counted 267 layoff events affecting roughly 185,894 workers, with about 56 percent of those events citing AI, automation, or machine learning as a driving factor. Those are the cuts where a company said the quiet part out loud. The real figure is almost certainly higher, because plenty of reductions get attributed to restructuring or efficiency instead.
Is AI causing mass unemployment?
No, and anybody telling you otherwise is selling something. Broad employment data does not show the collapse that was predicted. Most people still have their jobs and unemployment has not fallen off a cliff. What the data shows is narrower and more specific: white collar job openings are near their lowest levels in roughly a decade, and the cuts are concentrated in a particular category of work rather than spread evenly.
What is the approval-step layoff?
It is the pattern underneath the numbers. Look at which roles are actually being eliminated. Data entry. Bookkeeping and payroll clerks. Administrative assistants. Customer service. Entry level analysts and engineers. Every one of those is a job where a human being was the checkpoint between an input and an outcome. Somebody typed it in, somebody verified it, somebody signed off. When a company automates the checkpoint, it does not eliminate the work. It eliminates the person whose entire role was to be the checkpoint.
Which jobs are safest from this?
Work that cannot be reduced to a verification step. Anything requiring physical presence, licensed judgment with liability attached, relationship trust built over years, or accountability that a company cannot outsource to a classifier. The pattern is not manual versus knowledge work. It is whether your value is checking something, or deciding something nobody else can be held responsible for.
What is the oversight paradox?
Companies are cutting the very people who hold the institutional knowledge, while simultaneously lacking a trained workforce able to direct, validate, and quality-control the AI meant to replace them. IBM's own system now handles roughly 94 percent of routine HR tasks. That last stubborn percentage is where judgment lives, and the people who knew how to exercise it were in the group that got cut. Several companies are quietly rehiring for exactly that gap.
What should a young person entering the workforce do about this?
Understand that the bottom rungs are being pulled out, and stop trying to compete on execution speed. Entry level roles were how people historically built judgment: you did the junior work, you made mistakes, you learned what good looks like. That path is narrowing. The counter-move is to get close to accountability early. Work where decisions get made and consequences land, learn the tools well enough to direct them rather than compete with them, and build the kind of trust with real people that no system can produce on demand.
Connor T. MacIvor · CalDRE #01238257 · Sync Brokerage, Inc. · DRE #02031490