Back in My Day
You hear it at every tech conference. The veteran lawyer or judge steps to the microphone and eventually says, “Back in my day…” We all then start our own journey down memory lane with the speaker: frozen desktop computers, WordPerfect, jammed fax machines, manual bates-stamping, and the walk to the clerk’s office for a file-stamped copy.
After we have had a good laugh, we need to start imagining what our successors will say about us in twenty years. Someone will stand at the same podium and give their own “back in my day” speech about the era we are living through right now. That should make us pause, because we are the ones designing the next twenty years of the justice system. The question is not only what the tools will be able to do, but what kind of system we want to build with them. Just because something is possible does not mean it is wise, and just because it can be automated does not mean it should be handed over to the bot overlords.
Picture this: twenty years from now, someone will walk up to that same podium at a legal tech conference and say, “Back in my day, we still had to prompt the AI ourselves. Context windows were small. The systems hallucinated. And we had to verify every citation.” They will tell eager young associates and legal technologists about a world that still feels modern to us but ancient to them.
That future is not automatic, and it is not guaranteed to bend toward justice unless we decide what we want, not what the machine wants. Vision matters. Restraint may matter even more. So the question is not how far you can push each tool, but how far should we allow them to go without losing what gives the system legitimacy: human judgment.
In twenty years, most of the work will likely be done by agentic systems, AI that does not wait for prompts but initiates its own tasks, connected to court databases, capable of reading entire dockets, cross-checking every fact, and drafting opinions that cite, link, and reason in real time. Discovery will not be a battle of document requests and privilege logs. It will close when both sides’ models agree that all relevant information has been exchanged. Briefs will not cite yesterday’s law. They will update themselves in real time as new decisions come down. The record will not need to be assembled. It will assemble itself.
And that raises the question no one is really asking yet: What is left for us to do?
The easy answer is “judgment.” But judgment is not a single act. It is deliberation and interpretation, the weighing of competing values, the application of standards that resist reduction to code. A hearing is not legitimate just because the outcome is correct. It is legitimate because someone was heard, because reasons were given, because a person in a robe made a decision and owned it.
If the system writes the opinion, checks the record, and updates the law in real time, what work remains that cannot be delegated? And if we cannot answer that question clearly, we will delegate by default, not because we chose to, but because we never decided not to.
The danger is not that these systems become capable. The danger is that we fail to decide what must remain human. We need to draw those lines now, not in reaction to a crisis, but in anticipation of a world where the tools work so well that we forget why we ever did the work ourselves.
So yes, twenty years from now, someone will take the stage and say, “Back in my day, we had to use Claude and ChatGPT.” And the room will gasp, the way we used to when someone said they once used a typewriter.
But I hope that same person will be able to say, “Back in my day, we knew what was worth keeping.”
