New data point that stopped me cold: Uber burned through its entire 2026 AI budget in just four months, and its COO is now openly asking whether the spend is even worth it.
I kept seeing that figure pop up and had to dig into where it came from. Turns out it’s part of a sharp breakdown from Matthew Berman, the creator behind this video, who pulls together a pile of receipts to answer one question: is AI actually helping?
His answer flips the doom narrative on its head. Both Sam Altman and Dario Amodei have walked back their scary predictions. Altman called himself “pretty wrong” about AI’s near-term economic hit. Amodei, the guy who warned AI could wipe out 50% of white collar jobs, now says automation may actually expand the work people do.
📊 The insight breakdown
So if AI was supposed to gut entry-level jobs, why is the data going the other way? The original poster lays it out simply:
- Apollo’s chief economist says there’s “zero evidence” of AI-related job losses.
- US payroll numbers (ADP weekly) have been climbing right alongside AI spending.
- The big layoffs at companies like Block, Twitter and Meta? He argues those were bloated, over-hired teams from the zero-interest era using AI as a convenient scapegoat.
His core lens is Jevons paradox. When a technology gets cheaper, we don’t spend less, we spend way more, because suddenly use cases that never made financial sense become possible. More demand, more output, and yes, more humans needed to prompt and verify.
🛠️ 3 practical applications
Here’s how the creator suggests you actually act on this:
- Skip the frontier when you don’t need it. He points out most tasks don’t require the priciest models. Claude Opus runs around $25 per million output tokens, while options like DeepSeek land near 87 cents. Match the model to the job.
- Think “software factory,” not “more code.” He references Peter Steinberger spending $1.3 million in tokens in a month, not to write code directly, but to build a system that writes the code. Invest in the framework, not just the output.
- Become the AI-native person at your company. His advice is blunt: keep experimenting, learn what’s real versus hype, and you’ll stay valuable no matter what.
⚠️ Tips and pitfalls
- Watch the bottleneck shift. The expert makes a great point: you can ship a thousand features, but if nobody markets, packages or sells them, those tokens are wasted. The constraint moves to the rest of your business.
- Don’t buy “push a button, build a company” demos. He flags startups selling that vision as if it exists today. He likes the vision, but says there’s no proof it works now.
- Remember AI is great at “middle to middle” work, weak at end to end. A human is still needed on both ends, prompting at the start and verifying at the finish.
What I appreciated most is his honesty. He admits he’s not sure he’s still at the frontier of how people use these tools, and he’s actively rebuilding his own workflow to test what’s genuinely possible versus what’s just performance.
The takeaway I’m sitting with: this isn’t an AI bust, and it isn’t a magic jobs apocalypse either. We’re in the messy middle, figuring out how to actually diffuse the tech. Adoption is just slow.
Want the full set of receipts, charts and his take on whether we’re in a bubble? Watch the complete video, it’s worth the time.