May 2026 ยท 3 min read
AI's Free Buffet is over.
Why AI tooling is hitting the same wall Uber hit.
If you've ever flown business, and let's assume you didn't pay for it, you knew with the first clink of real glass champagne flutes that you were spoiled for life. No matter what life brings you afterwards, every time you're back to Ryanair's queues of crying students being charged twice the price of their ticket for their lead-dense backpacks, or easyJet's in-flight bingo, every single time, you go back mentally to that impeccable bloody mary a smiling, smart, charming flying attendant served you once while you were laying flat watching your third movie over the Atlantic.
Well, something like that is happening now with AI, where companies have been subsidising token-cost like there's no tomorrow, burning billions of dollars on your birthday invitation with at least one hidden typo, videos of you punching Tom Cruise in the face (that was worth every penny) and the like. All to get you hooked on AI. Mission accomplished (Half of US population has at least tried AI in the last year) they're throttling down the free buffet and increasing prices. Moving from the 20x, 30x they were covering down to 5x or so. Still a hell of a deal, only the buffet trays got smaller.
Companies like Microsoft, Uber and Duolingo have started to feel the pain and cutting back on "hey, go mad, burn tokens!" to "let's call it a day, chaps" and realising people, at least today, cost less and rarely go on doom loops.
Uber, precisely, is the standard bearer of tech's price dumping to lure customers, subsidising up to 60% of trip costs through most of the 2010s in order to undermine the evil cab workers and companies, entrenched in their monopolistic and abusive behaviour. Travis Kalanick famously framed it as a political campaign where "the candidate is Uber and the opponent is an asshole named Taxi". Once they did it, though, they started rising prices, applying dynamic price surges, got rid of Kalanik for being an asshole and became themselves a de-facto monopoly in very many countries.
AI's lack of a clear ROI has been argued as a reason to stop execs from burning $1K in a two hour presentation, or employees from spending tokens on games to rank higher in the AI employee of the month list. As any armchair LinkedIn expert such as myself, obviously knows, agentic tools burn ten to a hundred times the tokens of a chat session, which is the actual physics forcing the buffet to close.
Now, if you've refactored half a million lines of code in three days, ROI is crystal clear to you. Or if you've redone your 3K+ titles back catalogue in a day into epub3, with a consistent credits page, sorting all kinds of ISBN disasters, metadata, blurbs, socials and their mothers, that's been your AI Bloody Mary.
Soon doing that work will cost more than your salary, and therefore you're back to Excels and stuff, staring at the infinite, knowing this could be a perfect job for Claude to do while you lay back scrolling through IG's videos of Londoners on a sunny and hot Thursday afternoon, sipping (or guzzling) well-earned pints at the end of their at-the-office week.
I expect this will be just a temporary valley of death that will force everybody to build more efficient workflows, start looking at open source AI models that will soon be able to do what top models do today (ChatGPT 3 anyone?), and just like the '70s crisis ended up with smaller, more efficient cars, life will go on, and we'll get our tinier but still-delicious bloody marys on our desks, waiting for Thursday's bell so we can run to the pub like schoolkids at the gate.