AI arrived in public life as a novelty: a machine that could chat. Then it matured into something less cute and more unsettling: a machine that can act. Not merely answer questions, but pursue goals, take steps, and adapt to circumstances—an agent, not just a program.

That shift in vocabulary is more than a technical nuance. It becomes a fresh lens on an ancient subject. It forced me to revisit Judaism not as a museum of doctrines, nor as a ladder for climbing toward divine favor, but as something structurally different: a long-running system for guiding a human being through the moral and practical challenges of life in this world.

What AI did was not to “explain away” Judaism. It clarified its architecture—especially its governance architecture.

Programs follow rules. Agents pursue purpose under constraint. And that difference changes everything.

Programs and agents: the distinction that matters

Programs follow rules. Agents pursue purpose under constraint. And that difference changes everything.

A traditional computer program is essentially a recipe. It follows explicit instructions. If the environment changes, the program doesn’t reinterpret its purpose; it either fails or waits for an engineer to rewrite it. Programs do what they are told.

Agents are different. They are built around goals, not just steps. They operate in messy environments. They decide what to do next, often using their own judgment. And because goal pursuit in a complex world creates new failure modes, agents can “drift”—not because they have souls, but because optimization under uncertainty produces predictable problems: proxy-chasing, rationalization, blind spots, edge cases, and unintended consequences.

Once you see agents this way, a new concept becomes central: governance.

Programs are governed mostly before launch: requirements, tests, controlled deployment. Agents are governed continuously: ongoing constraints, updated training material, monitoring, auditability, and corrective feedback. In short, agents don’t just need code; they need a living framework that keeps them aligned.

And that makes an old truth newly legible: human beings do not operate like programs.

We interpret, choose, justify, regret, repair, and learn. We also drift, optimize proxies, and build elaborate stories to excuse what we already want. The modern agent language doesn’t reduce humanity. It illuminates it.

The agent context: what every agent needs—and what every agent risks

An agent is not just intelligence. It is a structured relationship between four things: