They Lied to Us About AI

Overview

Mo Bitar delivers a sharp, comedic critique of the AI industry's central narrative: that AI is about to replace everyone's job. He contrasts OpenAI's internal dysfunction and product sprawl with Anthropic's disciplined focus, then pivots to the evidence that current AI tools aren't replacing developers at all. The video builds to a pointed argument that the entire AGI narrative is a bait-and-switch designed to maintain hype and investment.

Key Concepts

OpenAI's Focus Problem

In February 2026, OpenAI hired Fidji Simo from Facebook and declared a "code red" because the company had too many projects running simultaneously. The irony: the company selling AI productivity tools couldn't stay productive itself, and solved the problem with meetings rather than AI.

Product Sprawl vs. Discipline

• OpenAI's product rampage: Sora (TikTok-like video app), Atlas (browser that MIT couldn't make sense of), ChatGPT Shopping (hallucinating ways to spend money), plus internal organizational dysfunction with teams placed under wrong divisions
• Anthropic's focused approach: Fewer people, less money, less compute, but focused on two things: coding tools and enterprise. Claude Code became so popular that developers used it over Christmas, with the Wall Street Journal calling it a "Claude Bender"

The METR Study: AI Made Developers Slower

A randomized controlled study by METR recruited 16 senior developers working on real codebases with real tasks. The finding: when using AI, they were 19% slower. While 41% of code is now AI-generated, the productivity gains aren't materializing. Bitar argues AI is "creating a second job for developers, not replacing them" because someone still has to review all the AI-generated code.

The AGI Bait-and-Switch

• Companies released LLMs and called them "AI," generating massive hype
• When people realized LLMs are sophisticated autocomplete (predicting the next word based on patterns), not actual intelligence, the industry rebranded the goal as "AGI"
• The $300 billion raised was based on calling a "fast bullshit generator" AI, then telling people the "real" AI is coming later
• LLMs cannot learn on the job; they can only do what a lab trained them on tasks generic enough to appear millions of times on the internet

Why Your Company's Work Is AI-Proof

Every company has unique internal tools, proprietary systems, and undocumented processes that require deep judgment and company context. This work doesn't exist anywhere on the internet for an LLM to train on. The thing that could replace your job (AGI) is imaginary, and the thing that's real (LLMs) can't replace your job.

Key Quotes

• "The company selling you the productivity tool got less productive. That's like a sleep doctor falling asleep during your appointment."
• "We never actually invented AI. We invented a very fast bullshit generator, called it AI long enough to raise $300 billion."
• "The thing that could replace your job, AGI, is imaginary. And the thing that's real can't replace your job. And the entire industry is praying you don't notice this difference."

Critical Analysis

Bitar's argument is deliberately provocative and one-sided. The METR study is real but limited (16 developers). The 19% slowdown finding has been contested and may reflect the learning curve of new tools rather than their inherent limitations. His framing of LLMs as "autocomplete" is technically reductive, though it captures a valid point about the gap between marketing claims and current capabilities. The strongest part of his argument is the observation about company-specific, undocumented knowledge being inherently resistant to AI automation.