Research vs production. The myth of AI in law.
They sell us the revolution and deliver souped-up libraries.
The major legal publishers have all taken the AI turn. There is no denying it, their tools are powerful. But by adding an AI layer on top of their databases, they are making a mistake: a “non-practitioner” reflex.
A search engine, even on steroids, is still a search engine
Practicing law is not about spending your life searching for case law or legal scholarship. It is a preliminary and necessary step, but only one part of the work.
You also need to produce an argument or an analysis, and verify the relevance of the arguments raised. Moreover, whether it is litigation or (informed) legal advice, searching for case law or legal scholarship is not relevant in every case. Sometimes, you “simply” need to analyze the facts and compare them to the law. You need to know the law to understand what can or cannot be done.
The missing link: legal engineering
Today, many lawyers are looking for the miracle tool that will do everything. The truth? The most powerful tool you already have: it is your legal reasoning.
Creating an AI agent capable of detecting a procedural defect in a summons is not computer magic. It is legal engineering.
The lawyer has this mental algorithm: “If date X exceeds deadline Y, then nullity.” The problem is that current “off-the-shelf” tools are not wired for your reasoning. They are wired for their content.
My advice before buying “turnkey” solutions
Rather than stacking expensive licenses that promise the moon:
Go back to basics: Start with generalist and accessible tools to understand the mechanics and workings of AI.
Test the logic: Try to have the AI accomplish a simple procedural verification task (not a research task).
Tame the beast: It is by understanding the strengths (speed of analysis) and weaknesses (hallucinations) of the technology that you will be able to build, step by step, the custom-made agent that will truly save you time.
I believe the future does not belong to AI that “knows everything” but to AI that “does” what you have taught it to do.