Your boss didn’t decide to mandate AI because he read a Harvard Business Review article over the weekend. He was sold to. And the people who sold him are doing the math you should be doing.
Let me walk you through the math. Then I’ll walk you through the Robert Downey Jr. line from Tropic Thunder that, I think, explains this entire industry better than any analyst report I’ve read in the last year.
The trillion-dollar problem
Here is the spending side of the AI economy as of mid-2026:
Gartner: $1.5 trillion in global AI spending in 2025, projected $2 trillion+ in 2026.
Big Tech capex (the eight largest hyperscalers): $427B in 2025, projected $562B in 2026, $637B in 2027.
McKinsey: the world needs to spend $5.2 trillion on data centers by 2030 to meet AI compute demand.
Nvidia’s Jensen Huang: $3–4 trillion in AI infrastructure by end of decade.
OpenAI alone: $852 billion valuation, $1T+ in committed compute partnership deals.
OpenAI H1 2025: $4.3B in revenue, $13.5B in net losses — roughly $2B burned every month.
Cumulative OpenAI losses through 2029: projected to exceed $140 billion.
Pause on that last one. A company is on track to lose 140 billion dollars over a five-year window. It is valued at almost a trillion. It is committing to spend another trillion on compute. And the headline product — the one you, the consumer, see — is a $20 chatbot.
Does this look like a business that intends to make its money back from $20 chatbot subscriptions?
The $20 fantasy
Here is what the consumer side of OpenAI actually looks like:
ChatGPT has roughly 800 million weekly active users. Approximately 5% of them pay anything at all. Of those who pay, the bulk are on the $20 ChatGPT Plus tier. OpenAI’s gross margin is 33%. Inference costs alone are projected to hit $14.1B in 2026. This year, OpenAI started rolling out an $8 ChatGPT plan to chase volume in price-sensitive markets like India and Southeast Asia, because at $20 the tier had stopped growing fast enough.
You don’t roll out a cheaper consumer plan if your consumer plan is the path to profitability. You roll it out when you’ve already concluded that consumer subscriptions will never pay back the infrastructure bill, and you’d rather have 122 million people locked into your product at $8 than 47 million paying $20.
Because the $20 subscription was never the business. It was the funnel.
The real customer is your CFO
Where the math actually works for OpenAI, Anthropic, Google, and the rest is not on your laptop. It’s on the ledger of the company that signs your paycheck.
OpenAI’s enterprise pricing is $25–$60 per user per month. A 10,000-employee company is several million dollars in annual revenue. They have over 9 million paying business users as of February 2026, up from 5 million the prior August. That number is growing far faster than the consumer side.
But even enterprise seats, multiplied by the entire knowledge workforce of the planet, don’t quite close a $1 trillion compute commitment.
What does?
The agent that does the job instead of the worker.
OpenAI has reportedly hired 100+ ex-investment bankers at $150/hour to teach its models how to do “junior banker grunt work” — IPO models, M&A decks, restructuring analyses. They are explicitly aiming at consulting, finance, and legal labor budgets, which they openly describe as “billions in current labor spend.” The 90th-percentile revenue scenario in their own forecast — the one that gets them to $70B ARR — explicitly requires “autonomous agents that can replace human workers in customer service, software engineering, and research.”
That is not a euphemism. That is the slide deck. Your job is the line item.
Why your boss suddenly cares whether you use AI
This brings us to the part of the story most people experience personally without seeing its shape.
Microsoft made it official last year. An internal memo from Julia Liuson, President of the Developer Division, told managers to incorporate AI usage into “holistic reflections on an individual’s performance and impact.” In her words: “AI is no longer optional. It’s core to every role and every level.”
Google, Meta, Shopify, Duolingo, KPMG, Thomson Reuters, JPMorgan Chase, Citi, and Amazon have all rolled out variations of the same policy. Amazon’s Forte system now produces an Overall Value (OV) score that explicitly tracks AI usage. Reports from inside Amazon describe a culture where “falling behind on AI means falling behind.”
Why the sudden urgency from leadership? Because the vendors that sold the seats need adoption to justify expansion, and your CIO needs adoption to justify the line item to the CFO. So you get a mandate. Use it. Document it. Justify yourself to it. Demonstrate impact. Or your performance score drops.
Notice what this is not. This is not your employer noticing a productivity gap and helpfully offering you a tool. This is your employer being told by a vendor that billions have been invested in a product, and the only way to recoup it is to enforce its use until it absorbs more and more of your work, until eventually it absorbs all of it.
The layoff receipts
If this sounds paranoid, it isn’t. The CEOs aren’t even pretending anymore.
Microsoft: 15,000 cut in 2025. Satya Nadella confirmed ~30% of the company’s code is now written by AI.
Intel: 15,000 in 2025, another 24,000 in early 2026.
Amazon: 14,000 in late 2025, another ~30,000 in early 2026. Beth Galetti pointed to AI as the reason the company could operate “more efficiently with fewer people.”
Accenture: 11,000 cuts. CEO Julie Sweet: “those we cannot reskill will be exited.”
Salesforce: customer support went from 9,000 to 5,000 heads. Marc Benioff says AI does “30 to 50% of the work at Salesforce now.”
Block: nearly half the company. Jack Dorsey: “the intelligence tools we’re creating and using are enabling a new way of working which fundamentally changes what it means to build and run a company.”
C.H. Robinson: 1,400 cut after rolling out AI pricing and shipment tracking.
Chegg: 45% of the workforce gone as students moved to ChatGPT.
Q1 2026 alone: 78,557 tech layoffs. 47.9% directly attributed to AI by the companies that did them.
Anthropic’s own CEO, Dario Amodei — the guy whose company makes Claude, which I’m partially using to write this — said in May 2025 that AI could eliminate up to 50% of entry-level white-collar jobs and push U.S. unemployment to 10–20% within five years. In January 2026, he doubled down in a 20,000-word essay calling it an “unusually painful” shock and warning that CEOs would “quietly stop hiring, then replace humans with AI the moment it becomes viable to do so.”
The man building the thing is telling you what it does. The Wall Street Journal called the latest wave “the week the dreaded AI jobs wipeout got real.”
The Tropic Thunder rule
Now we get to the cleverest piece of marketing the AI industry ever produced.
You’ve heard the line. Probably from a manager. Probably from a LinkedIn influencer wearing AirPods Max in a coffee shop. Probably last week:
“You’re not going to lose your job to an AI. You’re going to lose your job to someone who knows how to use AI.”
This is supposed to be reassuring. Don’t worry — you can still win, just learn the tool. It’s the soft version of the message. The hopeful version. The one that makes adoption feel like agency.
It is also, to borrow from Robert Downey Jr., complete nonsense the moment you examine it.
The line from Tropic Thunder:
“I’m a dude playing a dude, disguised as another dude.”
It’s funny because it’s a tower of recursive nothing. Each layer claims to be different from the next, but you peel them back and there is nothing inside. The character has no actual identity — he is performance about performance about performance.
The “you’ll be replaced by someone who uses AI” line is the same kind of tower. Walk it down with me:
You won’t be replaced by an AI. — Layer one. Reassuring.
You’ll be replaced by a person. — Layer two. Slightly less reassuring, but still human.
...who uses AI to do your job. — Layer three. Wait.
Stop there. Why does that person exist? What value are they adding that the AI alone isn’t? They’re prompting it. They’re correcting it. They’re feeding it your domain expertise, your edge cases, your institutional knowledge. They are doing reinforcement learning from human feedback in real time, on a salary, for a company that intends to keep that knowledge after they leave.
They are a dude prompting an AI, disguised as another dude doing your job. Until the AI doesn’t need them either.
The honest version of the sentence is:
You’re not going to be replaced by an AI. You’re going to be replaced by someone who trains the AI. And then they’re going to be replaced by no one.
Both versions of the line are predictions of the same end state. The “softer” version just buys an extra step of plausible deniability for the manager delivering it. It lets him keep a straight face while watching the headcount chart go down.
It’s a dude losing a job, disguised as a dude keeping a job, disguised as a dude using a tool.
The trainers are next, and we already have proof
If you think I’m being dramatic about the AI trainers being phased out — look at what’s already happened to the people whose entire job was training AI.
Scale AI was the company. The whole industry. Founded by Alexandr Wang. Powered self-driving training data. Powered RLHF for OpenAI, Microsoft, Meta, Anthropic. Built a global crowd of 700,000+ contractors through its Outlier subsidiary. At one point, “the human layer of AI” was effectively synonymous with Scale AI.
In June 2025, Meta paid $14.3 billion for 49% of Scale AI and pulled Wang in-house. One month later, Scale AI laid off 14% of its workforce — 200 employees and 500 contractors — and announced a pivot away from its core data-labeling business. Several of its biggest customers, including Google and OpenAI, walked at the same time.
Walked to what? Synthetic data. Teacher-models training student-models. The frontier labs have been steadily replacing the human RLHF layer with AI-generated training data and AI-judging-AI evaluation pipelines. Human trainers still exist for the hardest edge cases, but the trajectory is unambiguous: the labelers who fed the models are being eaten by the models they fed.
If the people whose literal job description was “train the AI” are being replaced by AI, the version of you that “knows how to use AI” should not feel particularly safe.
The play, in four moves
Strip away the marketing language and the entire AI economy is running a play that fits on a napkin:
Sell the worker the $20 subscription so they train the AI for free, every day, on their own work product.
Sell the boss the enterprise seat so AI does 30% of the work, framed as a productivity tool.
Sell the boss the agent so AI does 100% of the work, framed as cost optimization.
Quietly retire the AI trainers along the way, because synthetic data and self-play got good enough.
That is the plan. It is not a conspiracy. It is in the IPO filings. It is on the earnings calls. It is in Amodei’s interviews. It is in Benioff’s podcast appearances. It is in OpenAI’s own projected revenue model, where the high-end scenario explicitly depends on autonomous agents replacing human labor.
There is too much money invested in AI to let you keep your job. A trillion-dollar capex stack does not pencil out at $20 a month. It only pencils out at your salary, multiplied by everyone like you.
What this actually means
I am not telling you to refuse to use AI. That ship sailed. The tools are real, they are useful, and refusing to learn them is not a survival strategy — it’s just a slower way to lose.
I am telling you to be honest about the deal you’re being offered.
The deal is: train your replacement, on the clock, for free, while paying $20 a month for the privilege. In exchange, you may get to keep your job a little longer than the colleague who refused.
That’s a fine deal compared to refusing. It’s a terrible deal compared to building leverage that doesn’t live inside your employer’s HR system — your network, your reputation, your own products, your own audience, things you take with you when the AI catches up to whatever you currently do for a paycheck.
Use the tools. Of course use the tools. Just stop pretending the tool was sold to help you. It was sold to replace you. You’re paying the entry fee.
Don’t be a dude using a dude, disguised as another dude.
Be the one with the leverage.
If this resonated, share it with someone whose performance review now includes “AI usage.” They’ll either thank you or argue with you. Both are useful.