The only Permanent Underclass are the ones who believe it is permanent (www.swyx.io)

The piece argues that AI is accelerating wealth and power concentration—big labs and elite teams are pulling away (examples: Meta FAIR researchers making ~$500k vs >$25M at a hypothetical Meta “SuperIntelligence” lab; OpenAI’s $1.5M employee bonuses and $6B in secondary sales; Anthropic jumping from $4B to $183B)—and that this will amplify housing and ownership gaps in places like San Francisco. It also predicts a skills bifurcation: “agentic” coders who master coding agents and toolchains will dramatically outpace traditional coders, creating high-productivity tiny teams versus unhappy enterprise ticket-pushers. These are real signals that inequality can widen under capitalism and the API economy. But the author rejects the fatalistic idea of a permanently trapped underclass: agency matters. Faster, cheaper learning (English, coding, ML/AI engineering), easier model finetuning and (near-)frontier RL experimentation, remote work, and startup infrastructure (Stripe Atlas, YC) lower barriers and increase class mobility. The practical implication for the AI/ML community is clear—technical adoption and tool fluency (especially agentic workflows and model finetuning) will be the key dividing line, not raw IQ. In short: AI can both widen gaps and create new escape routes; outcomes will hinge on who learns and applies these tools with high agency.