Judgment is what the signature warrants.

A filing goes out under a lawyer's name. Rule 11 attaches to it, and in discovery, Rule 26(g) signature responsibility attaches to the certification. The signature warrants something specific: that a licensed professional made the calls that required professional judgment. The question automation raises for lawyers is therefore not which tasks a machine can perform, but which judgments it must not quietly take over.

In opposition practice we count six: the theory of denial the court must adopt for the motion to fail; the posture of the response, aggressive or surgical; each concession; whether each distinction actually holds, since materiality is a legal call rather than a similarity computation; reliance on each authority the opposition itself cites; and the final attestation. Everything between them, from parsing the motion to checking the gates, is execution, and execution is what machines are for. The six judgments arrive in a queue with their reasoning, provenance and draft, and each resolves the same way: the attorney adopts, rejects, or overrides with a stated reason, and the choice is logged.

The same line runs through our other work. In the multi-agent experiment, decisions a court would hold the company accountable for sat behind red lines reserved for humans. In the blind eDiscovery review, attorneys approve rubrics, handle privilege, validate production and keep signature responsibility.

This is what automation changes about lawyers, and what it does not. The mechanical share of the work shrinks. The share that is judgment becomes more visible, more concentrated and more consequential. A system that silently absorbs those judgments has not automated the work. It has transferred the malpractice, and made the lawyer sign for decisions the lawyer never saw.