Episode Details

The word “agent” comes from the Latin agere — to drive, to set in motion, to do. Its present participle, agens, meant “the one doing.” Not the one being done to. Not the one being managed, surveilled, optimized, or replaced. The one doing.

Hold that in your mind for a moment while I tell you what happened this week.

Forbes ran a piece titled “The Agentic Enterprise Doesn’t Mean Axing Half Your Staff.” The subtitle is meant to be reassuring. Daniel Dines, CEO of UiPath, and his CMO Michael Atala paint a picture of layered AI agents — task agents, stage agents, process-level agents — coordinating work across entire enterprises, with humans “supervising, guiding, and intervening where judgment is required.” Jensen Huang is invoked as the hopeful counterpoint to Jack Dorsey’s layoffs at Block: imaginative companies, Huang says, use AI to do more, and leaders who default to layoffs are simply “out of ideas.”

All of this is pleasant. None of it is serious.

I say that because in two thousand words about agents, agency, and the future of the enterprise, not a single sentence asks a question that any first-year Latin student would think to ask: Agent — for whom? To what purpose? And who had the agency before you automated it away?

The etymology is instructive. When agens entered English in the fifteenth century, it meant “one who acts.” By the sixteenth century, it had already drifted to “one who acts on behalf of another.” A deputy. A representative. Someone authorized to exercise power that belongs to someone else.

That drift — from doer to deputy — is the entire history of labor in a single word.

A worker on an assembly line is, etymologically, the original agent. She is agens — the one doing. She drives, she sets in motion, she performs. Her hands hold institutional knowledge that no org chart captures. Her judgment catches the defect that the sensor misses. Her intelligence — purchased, deployed, and then systematically ignored for decades — is the most stranded asset on any balance sheet in America.

But in the language of the “agentic enterprise,” she has been demoted from agent to patient. The one acted upon. The AI is now the agens. She is what it acts on, or around, or instead of. Her intelligence didn’t fail. It was never asked.

This is not a semantic complaint. It is an ontological one.

• • •

Dines makes a charming argument about “taste.” LLMs, he says, are “the average of everything,” and taste requires a body, experiences, a personality. Therefore humans still matter.

With respect: this is the argument you make when you’ve already conceded the architecture. If the best case for human beings in the enterprise is that we have taste — some vaguely aesthetic, irreducible je ne sais quoi — then we’ve already agreed that the real work belongs to the machines. Taste is decoration. Taste is what you hang on the wall after the building has been built by someone else.

I’ve managed manufacturing plants for thirty-six years across three continents. What I’ve seen on the floor is not taste. It is intelligence — specific, operational, earned, and breathtakingly precise.

Let me tell you about Jamie.

Jamie was one of my team leaders at GM’s Lansing Delta Township assembly plant. We were running nearly fifty-five jobs an hour — a car a minute, give or take. When a defect appeared half a mile down the line, in an entirely different section of the plant, Jamie would be moving before most people had read the alert. He knew what the defect was. He knew what tools to bring. He could walk it back a hundred cars while the line was still running, tracing the fault to its origin without stopping production, without sacrificing quality. To watch him work was to watch a mind operating at full deployment — reading the system in real time, making decisions faster than any algorithm