In a March 11, 2026 blog post titled "Every minute you aren't running 69 agents, you are falling behind" — the title a deliberate parody — George Hotz, better known as geohot, pushed back against what he called the toxic wave of <a href="/news/2026-03-14-against-vibes-evaluating-generative-model-utility">AI productivity panic</a> dominating social media. The comma.ai founder argued that the breathless rhetoric urging people to orchestrate dozens of agents or risk obsolescence is "complete nonsense," and that AI remains fundamentally "search and optimization" operating within well-understood theoretical limits. Hotz dismissed the notion of AI "going recursive" or achieving some transformative self-improvement loop, noting that search and optimization have always been recursive. Tools like autoresearch, he conceded, are impressive — but because search is impressive, not because AI is magic.
A substantial portion of the post addresses the economic anxiety driving the discourse. Job losses attributed to AI are more accurately explained, Hotz argues, by large platform players consolidating rent-seeking roles — jobs that create complexity or friction rather than genuine value. Companies cite AI as the reason for <a href="/news/2026-03-14-meta-weighs-20-workforce-layoffs-to-offset-rising-ai-infrastructure-costs">layoffs</a> because doing so boosts stock prices, but the underlying dynamic is winner-take-all consolidation in zero-sum competition. His prescription: exit zero-sum games entirely, create more value than you consume, and ignore social media comparison traps. At comma.ai, he has applied this directly — openpilot has been open-sourced from day one, and the comma four device sells for a flat $999 rather than via subscription.
That value-creation argument carries more weight given what Hotz has actually shipped. Openpilot, comma.ai's open-source driver assistance platform, has logged over 300 million real-world miles across more than 20,000 active users and supports more than 300 car models, according to comma.ai's public figures. He has also spent five years building tinygrad, an independent deep learning framework with 31,600 GitHub stars, explicitly designed to resist the abstraction creep he criticises in the agent hype cycle. The credentials aren't a preamble — they're the argument. Someone selling a $999 device and open-sourcing the software has different incentives than someone monetising AI anxiety.
The post generated mixed reaction on Hacker News. Some commenters agreed, noting that AI-generated generic content remains uncompelling and genuine value creation still matters. Others pushed back harder: one flagged apparent internal contradictions in Hotz's own post, while another made the sharpest point of the thread — that advising people to ignore financial returns is only realistic advice if you're already financially secure. Hotz's philosophy may be sound; whether it's accessible is a different question.