Interview · Three questions
Should a legal AI be able to do anything?
Tomer Libal, CEO of Enidia AI, asked our CTO Aleksander Smywiński-Pohl three questions about where AI agents in law are going. Here are his answers.
Everyone talks about AI agents that can do anything. In law, is that a feature or a liability?
It depends on what we acknowledge as a legitimate action. A deployed system has safety measures, but an agent without the full context can take an action that looks harmless on paper, like sending an email with a question to an employee. Legally, it may breach an NDA or GDPR if the question reveals information that person should not have. I think we will see many such incidents as autonomous agents spread.
In three years, will law firms buy AI from a few global vendors, or will most run their own?
Nobody knows, but it hinges on hardware prices. If the sharp rise in GPU and memory prices continues, more customers will move to the cloud. On the other hand, newer models need less compute and can replace the largest ones in many use cases. The one thing I’m sure of is that hardware prices are too high right now.
AI can pass exams designed for humans. Does that tell a lawyer anything?
Not necessarily. The real problem is who will read what an artificial lawyer generates. We already see agentic overproduction: agreements come back with a swarm of mostly useless comments, and litigation files fill up with hundreds of pages of superficial arguments nobody can process efficiently. It doesn’t matter if AI passes the exam. It matters if it simplifies legal matters. I would never buy a robot that generates more mess while cleaning.
The test we hold Enidia AI to: checking the AI’s work must be faster than doing the work yourself.
Try Enidia AI for free with your Google login.
Open app.enidia.ai