Oracle cut 21,000 jobs and pointed at AI. For most of us, the lesson runs the other way.

Oracle shed about 21,000 roles this year and named AI in the filing. If you run a mid-market business, copying that logic is the mistake. Your AI dividend is not a smaller payroll, it is the same people reaching work they could never get to.

Oracle's latest annual report has a number in it that is easy to miss and hard to forget. The company employed about 141,000 people at the end of May. A year earlier it was around 162,000. Somewhere in between, roughly 21,000 jobs went, and the filing is unusually blunt about why : the "deployment of AI technologies across our operations" has "resulted, and may continue to result, in reductions to our workforce".

That is about thirteen staff in every hundred, and roughly US$1.8 billion in severance and restructuring along the way. Oracle is not alone. Amazon and Meta have made their own cuts while pouring money into AI infrastructure, and by the layoff trackers' count, well over 100,000 tech roles have gone already this year. The story writes its own headline. AI arrives, people leave.

I want to be careful here, because that headline is doing more work than the facts quite support. And because, if you run a mid-market business in Australia or New Zealand, it is the wrong lesson to copy.

AI is a tidy reason for cuts that have other causes

A company that hired hard through the cloud boom, and is now spending tens of billions on data centres, has plenty of reasons to trim payroll. "We are deploying AI" is a cleaner thing to tell the market than "we over-hired" or "we are funding a building programme". That is not cynicism, it is how large public companies talk. The AI is real in those numbers. It is also a convenient frame. The difference matters, because the framing is what travels, and the framing is what people copy.

You are not Oracle

Oracle's problem is that it has more people than its new shape needs. That is not the problem most mid-market operators have. The businesses I work with are not sitting on idle capacity. They are the opposite. A finance team that has not closed the books on time since March. A service desk that never reaches the proactive work because it is buried in the reactive. A sales team doing its admin at night. There is no spare headcount to release. There is a backlog of work nobody can get to.

That changes what automation is for. If your team is already stretched, the win from AI is not a smaller payroll. It is the same people finally reaching the work that has been sitting in the too-hard pile: the analysis, the follow-up, the customer who quietly slipped through. You do not end up with fewer people. You end up with a team that covers more ground, and a few jobs that are less miserable to do.

The cut-first instinct is expensive

There is a practical trap in the layoff framing too. Cut the headcount first and automate second, and the thing you have actually lost is the knowledge of how the work was done. The person who left was the only one who understood why the month-end reconciliation has that one strange step. Automate around them and the step quietly breaks. (I wrote recently that AI cannot automate a process you cannot explain. The people who can explain it are often the ones on the redundancy list.)

The order that works runs the other way. Automate the dull, well-understood parts first. Watch where the time actually goes. Let the team tell you what they would stop doing if they could. Only then do you have an honest picture of what the work needs, and it is rarely "the same output with fewer humans". It is usually "more of the work we keep failing to get to, with the people we already have".

Oracle is solving an Oracle problem, in public, with a phrase that sounds like strategy. Borrow the tool if it is useful. Do not borrow the headline. For most of us, the point of automating the work is not to need fewer people. It is to stop wasting the ones we have.