Not more context, but better context (podcast)
An hour-long conversation about what actually happens when AI searches case law and legislation. Why ordinary models invent case numbers, how to prevent it, and what stays with the lawyer.
We recorded the podcast for a simple reason. There are plenty of bold claims about AI in law and very little explanation. We wanted to sit down in front of a camera and show, without shortcuts, how a system like this works inside, who is behind it and where its limits are. That is why the people building DirectCase are the ones on camera.
The most interesting part is the one that runs against intuition. Give the model more documents and the answer does not improve — often it gets worse. What decides is the quality of the context: the system has to find the right documents first, including the links to overruled case law, and only then write the analysis. We also talk openly about price, about client data security, and about why the machine searches but the lawyer decides.
What you will hear in the video:
- Why ordinary models invent statutory provisions and case numbers
- What happens under the hood when you ask a legal question
- Why more context hurts, and how the system spots overruled case law
- Price, tokens and why a monthly subscription makes sense
- The machine searches, the human decides: where AI ends and the lawyer begins
- What lawyers say after the first 10 minutes of testing
- Is legal reasoning at risk of becoming homogenised?
- DirectCase as a connector for Claude
- Client data security and when a local model makes sense
You can watch the full conversation on YouTube.