In an essay the authors rate at about 25% confidence, copying is the first lever: training cost is amortized across thousands of instances, justifying deeper expertise per AI and letting small successful teams be replicated across a thousand projects.
The essay argues copying changes management more than labor: copies of an AI CEO could set every product's strategy, review every pull request and run negotiations from one vision, with no principal-AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → problem.
Under 'merge', a central model would absorb what its distillationTraining a small, cheap model to imitate a big one's outputs, keeping much of the capability at a fraction of the cost.Full definition → copies learn through explicit summaries, shared latent representations or direct weight edits, with instances communicating through latent representations.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
distillation — Training a small, cheap model to imitate a big one's outputs, keeping much of the capability at a fraction of the cost.
Why it matters
Dwarkesh Patel, with Epoch AI's Ege Erdil and Tamay Besiroglu, argues that AI firms will gain speed from copying trained expert-level workers arbitrarily, a structural advantage distinct from raw model IQ that changes how automation should be modeled.
