On 16 October, Anthropic published a report about what AI did to their own company. Headcount, hiring, meetings, who gets interviewed and why. I read it twice. The second time with a glass of wine, because the first time was not fun.
Then I closed it and thought: this is not about them. What they see inside their walls today, you will see inside yours. Not in ten years. In your next two budget rounds.
So here are my six predictions. No research, no footnotes. Just a coach who has spent ten years in rooms where leaders tell me what they really plan to do, as opposed to what they put in the town hall deck. :)
1. Junior jobs disappear - and no one gets fired
There will be no headline. No tearful post from a 24-year-old who lost her job.
Your company will grow. The junior line in the budget will stay flat. Next year someone merges it with another line “for simplicity”. The year after, it’s gone, and the only person who remembers it is the controller.
At Anthropic, junior tech hires went from a quarter of all hires to under a tenth. While the company tripled. That’s the pattern. Growth, and quietly not hiring the people who used to do the first draft.
I’ve written it maybe fifty times in 500 articles and I’ll write it again, because the soft version is still winning: AI takes jobs. The first ones it takes are the ones where you learn the profession by doing the simple parts. The model adores the simple parts.
2. Your best people become the bottleneck. And they will hate it.
Picture your sharpest senior. The one who actually gets things done while everyone else is in a workshop about getting things done.
In two years she has ten agents working for her. They produce more than she can read. Her job is no longer doing. Her job is reading, approving, correcting, approving. One engineer in the report called it babysitting: every agent comes to you with the same panic, whether the juice box is stuck or someone is bleeding.
Hmm.
What she won’t be told: this is a management job. She got promoted without a title, without money, without one day of training. I’ve coached this person already, by the way. Three of them this year. The symptoms are the same: shorter fuse, longer days, and the sentence “I didn’t do anything today, I just answered things.”
That’s not burnout from work. That’s burnout from responsibility with zero support. Classic. We’ve only had 40 years to learn it.
3. You will have more meetings. Sorry.
Yes, you were promised the opposite. I was in the room when the vendor promised it.
But think it through. The machine fixes the bug in four minutes. What’s left is: why was the bug there? And the answer is always a people answer. Two teams, two truths. A decision made in a corridor. Someone who said yes to be nice.
Machines are brilliant at doing. They are useless at sitting between two directors who hate each other. Anthropic’s meeting time went up 20% in a year. That’s an engineering company. Imagine a bank.
So if your meetings are bad today - and they are, I’ve sat in them - they are about to become your biggest competitive weakness. Fix them now. Not with a tool. With a leader who is willing to say “this is a decision, not a discussion.”
4. Your interview questions expired last spring
Most hiring still tests whether the candidate can do the thing. Write the code. Draft the contract. Build the deck.
The model can do the thing. Faster than the candidate, and it doesn’t ask for a parking spot.
The hiring managers at Anthropic have already moved on. What grew in their interview notes: judgment, independence, “uses AI in a clever way”. What shrank: expertise, execution.
I expect this in every function in two years. The 2027 interview has three questions. What would you decide here? How would you notice if you’re wrong? Show me something you built with an agent for a real problem, not for LinkedIn.
If your scorecard has no line for judgment, you’re hiring for a job that is being deleted while the candidate waits in reception.
5. Transparency stops being a poster and becomes a system requirement
This one surprised even me, and I’m hard to surprise after six WorldSkills.
Agents only know what’s written down. If a decision is made over coffee and never logged, every agent in the company keeps working on yesterday’s truth. One guy in the report kept getting t-shirt requests for months because his name was on an old document.
So companies will be forced to move human work into the open. Decisions in writing. Feedback in channels, not in the kitchen. Leaders defending their choices where everyone can read them. Including the machines.
For most organizations I know, this is the bigger shock. Not the technology. The honesty. Try telling a Hungarian middle manager that from now on, the reasons behind his decisions are public. I’ll bring popcorn.
6. “Who owns my knowledge?” becomes the fight of the decade
Your people are now typing their expertise into instruction files so a machine can use it. Ten years of judgment, reduced to a prompt. For free.
Anthropic handles it with stock. Everyone owns a piece, so teaching the agent is teaching your own pension.
Most of you can’t offer that. So what do you give the person who spends six months teaching an agent to do 40% of her job? If the answer is “the opportunity to grow”, I promise you she will grow. At your competitor.
Some way of sharing the gain is coming. Bonus for automating yourself. Four-day week. Something. The ones who figure it out first keep the people who matter. The rest keep the people who didn’t bother.
And the part no company can fix
Anthropic writes it honestly, and I respect that: everything they do protects the people already inside. It does nothing for the 22-year-olds who never get in.
That’s where I land too. HR won’t solve this. Governments will have to, and they haven’t even read the report. Until then, the smart young ones will stop waiting for an entry-level job and walk into the interview with an agent already built. The rest will wait. Waiting is not a strategy.
What to do next
- Leaders: count how many of your seniors already manage machines without a mandate. Give them one. Title, training, time. This week.
- HR: put judgment and AI fluency on every scorecard. Remove one skill a model already does. Yes, that one.
- Everyone over 35: your value is your judgment, not your output. Say it out loud in your next review. Then charge for it.
- Everyone under 25: build one agent for one real problem. Bring it to the interview. That is the entry-level job now.
The first step of a career is disappearing. Who are you going to pull up?