AI legal process automation still runs into the same review ceiling
Every argument for AI legal process automation ends where every argument for offshore hiring ends: someone has to read the output. A drafting tool that produces ten first-pass demand letters in an afternoon has not given the firm ten letters. It has given one attorney ten things to check, and that attorney's day did not get longer. The tool moved the work; it did not move the ceiling.
Why AI legal process automation does not add capacity on its own
Production was never the scarce thing in a 30-attorney firm. Judgment was. Intake summaries, deposition digests, discovery responses, first-draft correspondence: these are outputs a competent tool can generate at volume, and the volume is not the constraint. The constraint is the number of hours a licensed attorney has available to decide whether a given output is correct enough to send.,
That constraint has a floor set outside the firm. The supervision a lawyer owes over delegated work is a professional obligation, and the ABA Model Rules describe it in terms of the lawyer's responsibility for the result rather than the method that produced it. Faster production does not dilute that responsibility across more units. It concentrates more of it into the same day.
So the honest accounting of an automation project is not hours of drafting saved. It is hours of drafting saved minus hours of review added, and the second number is frequently larger, because reviewing something you did not write is slower than reviewing something you did. The firm buys speed at the front of the pipeline and pays for it at the narrow end. That is not efficiency. That is a queue with a nicer entrance.
The firms that get value from automation are the ones that could already delegate
Grant the case for moving fast here, because it is a real one. Rates are under pressure, clients are comparing quotes, and a partner who waits for perfect conditions before adopting anything will be outrun by one who does not. Every practical improvement a firm has ever made started before it was fully ready. Adoption under uncertainty is not recklessness, it is how operations improve at all. But readiness is not a mood, and it is not a budget line. It is whether the firm has written down how the work is supposed to go.
A process that lives in one senior paralegal's head cannot be handed to a tool, because there is nothing to hand over. What can be handed over is a documented sequence with a defined output and a named person who checks it. Firms in that state can automate a step and see immediately whether the step got better. Firms without it are automating a guess and grading it by feel.
A low rate for unreviewed output is expensive. Free output nobody reads is more expensive still.
This is the same test that decides whether an offshore hire lands, which is why the two decisions should be made together rather than in separate meetings. Both add production. Neither adds a reviewer. A firm running both at once, on the theory that one is technology and the other is staffing, has doubled the pressure on the single resource it never expanded.
Buy review capacity first. Then automate into the room it makes.









