Managing Machines vs. Managing Humans. What's The Difference?

Simon Shaolei Du proved the math behind why deep learning works — the first global convergence result for gradient descent on deep networks — then left the whiteboard to run 100 AI agents at once. His claim: a large model is less random than a person, so 100 agents are easier to manage than 100 people. At Apodex he builds a deep-research "solver" where a separate agent's only job is to distrust the others — checking sources, reasoning, and code before any answer ships. Here's how a multi-agent team beats a bigger model, why a 10-million-token problem doesn't need a giant model, and what he tells CS grads who can't find a job.