Machine Proxies

Chapter 1: The Delegation Instinct

On a Tuesday morning in the autumn of 2023, a lawyer named Michael Cohen walked into a federal courtroom in Manhattan and asked the judge for permission to use an AI chatbot to review millions of documents in a class-action lawsuit. Cohen, who had spent decades as a personal attorney and fixer, told the judge that the cost of human review—associates billing at $400 an hour, paralegals with their highlighters and their binders—would be prohibitive. He proposed instead to let a large language model do the reading. The judge, a pragmatic man named Jesse Furman, asked a few pointed questions. Then he agreed, with conditions: the AI’s work would be double-checked by human lawyers, and any mistakes would be Cohen’s responsibility. “The technology is here,” Furman said from the bench. “We have to adapt.”

That scene, unremarkable in a city where lawyers routinely bill their clients for the privilege of reading other people’s emails, was nonetheless a quiet landmark. It wasn’t that a machine had been allowed to help with legal work—that had been happening for a decade. What changed was the authority. Cohen wasn’t asking the AI to suggest relevant cases or flag suspicious text. He was asking it to stand in for the lawyers themselves, to make thousands of micro-judgments about relevance, privilege, and evidentiary weight. The AI would not be a tool. It would be a proxy—a stand-in that spoke, decided, and acted on behalf of its human principal, with the