The Costs of AI Acceleration

Summary

The essay describes how its author rebuilt studying and software development around AI agents, using Omarchy and Gnosis to turn lecture material into summaries, custom study sessions, anatomy models, explanations, and continual quizzes. An agent can produce a lecture summary in about two minutes, compared with the three to five hours a colleague previously needed, and the author says this change coincided with a move from average grades to being among the top students. The productivity gain has weakened the practical need for study collaboration: coordinating notes with classmates now feels slower, although the author still values their company and recognizes the need to build relationships that do not depend on shared tasks. Studying has also become difficult to stop because the agent creates a continuous, interactive conversation and makes improvements to the study tools immediately usable. The author says agents have generated millions of lines of code across languages, making agent management more valuable to this workflow than writing all the code by hand. That efficiency comes with a new dependence on computing hardware: owning software does not provide the machines needed to run powerful agents, and the author estimates that work requiring $100,000 in compute today would ideally become possible for $5,000. The essay supports broader access to agent-assisted creation but argues that genuine self-reliance ultimately requires affordable chips, manufacturing capacity, and compute owned by users rather than rented from a few providers.