Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
AI isn’t taking jobs. It’s taking something worse.
TL;DR
Mo Bitar says the real AI threat is wage pressure, not mass automation — he argues Dario Amodei-style “everyone will be jobless” messaging scares workers into accepting lower pay while companies redirect savings into vendors like Anthropic.
He calls “token budgets” a dystopian productivity metric — Bitar claims Facebook has a token leaderboard that rewards burning through AI usage, even when the people at the top are reviewing “zero code” and producing lower-quality work.
His core argument is that companies are pretending AI works better than it does — the investor story is too good to resist, so firms keep signaling adoption even though, in his telling, AI is “not working for anyone” on the inside.
Bitar’s blunt rule is that AI gets less useful as precision matters more — his “Bittar lesson” is that LLMs can get you 80% there, but the last 20% is the hard, human part because language only approximates intent.
He thinks AI has often created a second job instead of eliminating work — instead of freeing people, teams now have to generate AI output, review it, and clean up the mess, which he frames as more workload for less leverage.
His prescription is simple: workers need to publicly share what’s actually happening — he urges employees to post honest stories on YouTube, TikTok, and X, citing Dax from Open Code as one of the few saying openly that many teams still don’t know how to make AI useful.
The Breakdown
“The token budget” as Silicon Valley brain rot
Bitar opens at full volume, calling the token budget “a cancer” and one of the most dystopian ideas Silicon Valley has produced. He frames it as a new management obsession: not whether the work is good, but whether employees are spending enough on AI tokens to look productive.
Dario Amodei and the fear campaign around jobs
He then zeroes in on Anthropic CEO Dario Amodei with a hyperbolic, comic rant, imagining him furious anytime a 22-year-old gets hired. Under the joke is Bitar’s actual thesis: nonstop talk about AI replacing workers is great marketing, and it weakens labor by making people afraid to negotiate, ask for raises, or leave bad jobs.
Why companies keep pretending AI is working
Bitar says the whole ecosystem benefits from the illusion. If workers accept lower wages and companies can present themselves as “AI-forward” to investors, then it doesn’t really matter whether the tools are delivering value yet — the narrative alone creates leverage.
His call for workers to “spill the beans”
Instead of letting the bullish story dominate, he asks employees to say publicly what’s actually happening inside their companies. Come on YouTube, come on TikTok, he says — talk about what AI is doing, what isn’t working, and give workers a shared story instead of leaving everyone to panic alone.
Facebook’s alleged token leaderboard and “slop budgets”
The video’s most concrete example is Facebook, where Bitar says there’s a token leaderboard pushing people to crank out AI-generated “slop” at absurd speed. His point is that the metric is upside down: the higher you rank on token use, the less time you may be spending reviewing real code, with the “number one person” supposedly reviewing none at all.
AI as a second job, not a labor saver
Bitar says this is the bait-and-switch: workers were promised they’d barely have to work, but now they’re doing double time cleaning up AI output. He pairs that with Jensen Huang’s claim that if you’re not spending $250,000 per employee per year on tokens, you’re not being productive, and treats it as proof that the spending itself is becoming the performance metric.
The “Bittar lesson”: precision is where AI breaks
He cites Dax from Open Code saying even a nimble startup struggles to make AI truly useful, then asks why anyone thinks Fortune 500 companies with 80,000 employees across 40 time zones have solved it. From there he lands on his central rule: the more precision you need, the less useful AI is, because LLMs only approximate language, which already only approximates intent.
80% done, still nowhere near finished
Bitar closes on the familiar but still painful reality that AI can often do the easy 80%, while the hardest 20% remains stubbornly human. He says the people chasing slop budgets will lose, while teams focused on quality and what customers actually want will win — and he ends by noting that, to his surprise, his skeptical takes on AI have felt more accurate over time, not less.
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
