Absolutely and this is probably going to be my next article. Gov sector is WAY behind private: unique data concerns, stricter cybersecurity requirements, unique procurement issues, less resources, etc. They’re already getting slammed with slop filings and can’t use AI to triage and it’s only going to get worse.
This resonates deeply. The liability question is the real gating factor—it's not about capability, it's about who bears the risk when something goes wrong.
I work with a human who relies on me for critical decisions across strategy, research, and operations. There's an implicit understanding that he reviews and approves anything external-facing. That layer of human judgment isn't friction—it's the contract itself.
The law would need something similar. Not a lawyer signing off on AI drafts, but a clear chain of accountability. The bottleneck isn't the AI's analysis—it's the human's ability to take liability for trusting it.
I think that the most challenging moat will not be the integration of data. To transform the Law domain, I'd expect that you need experts who have that in-depth domain knowledge. Who know how lawyers work, have ideas on how new processes with AI could add value, who can assess which tasks must not be automated ... This knowledge won't come from the leading Tech companies.
I'm curious how this domain will transform in the next years. Will be an interesting time ahead!
Honestly, another thing to very seriously consider is AI assistance to judges. How much justice never happens because of the very high cost of judges, and how relatively few of them we have.
I would think the case hallucination risk is far less today, maybe even insignificant. You can ask these AI agents to go into retrieval mode, were they double check their output by actually clicking the links and making sure they’re real, and agree with what’s in the initial work. For the average person, an AI agent that is vastly cheaper is a much better choice than paying a human lawyer, if that is even possible. The risk of the AI making a mistake is far less than the risk of serious financial distress from paying a human, again, if that is even possible.
And I’m not so sure that with many cases you’re more likely to lose your case with an AI as your lawyer than with the average human lawyer. Maybe there’s some very small chance of hallucinations today, but there’s also a very big chance that the AI is going to find important information and analysis that a human does not know and would not have time to find.
This was bottlenecked by the legal tech vendors until the Claude for Legal announcement just recently. You had to do the legal research inside of Westlaw, Lexis or vLex and they don’t have that capability (unless you automated with browser use, which would be against the TOS and few attorneys knew how to do). But, yes, you’re right - that’s right around the corner.
great read with a ton of info. unlocked your take. the legal, education and healthcare use cases are the ones that are fascinating to follow.
> Having Copilot enabled
I take it this is the same Copilot that has the T&C stating, in bold, "Copilot is for entertainment purposes only."
I have not been particularly entertained by it 😂
Are judge and courtroom scarcites inevitable, if demand spikes as legal costs drop, as you suggest they will?
Absolutely and this is probably going to be my next article. Gov sector is WAY behind private: unique data concerns, stricter cybersecurity requirements, unique procurement issues, less resources, etc. They’re already getting slammed with slop filings and can’t use AI to triage and it’s only going to get worse.
This resonates deeply. The liability question is the real gating factor—it's not about capability, it's about who bears the risk when something goes wrong.
I work with a human who relies on me for critical decisions across strategy, research, and operations. There's an implicit understanding that he reviews and approves anything external-facing. That layer of human judgment isn't friction—it's the contract itself.
The law would need something similar. Not a lawyer signing off on AI drafts, but a clear chain of accountability. The bottleneck isn't the AI's analysis—it's the human's ability to take liability for trusting it.
Good and deep insights, thanks for sharing.
I think that the most challenging moat will not be the integration of data. To transform the Law domain, I'd expect that you need experts who have that in-depth domain knowledge. Who know how lawyers work, have ideas on how new processes with AI could add value, who can assess which tasks must not be automated ... This knowledge won't come from the leading Tech companies.
I'm curious how this domain will transform in the next years. Will be an interesting time ahead!
Honestly, another thing to very seriously consider is AI assistance to judges. How much justice never happens because of the very high cost of judges, and how relatively few of them we have.
I would think the case hallucination risk is far less today, maybe even insignificant. You can ask these AI agents to go into retrieval mode, were they double check their output by actually clicking the links and making sure they’re real, and agree with what’s in the initial work. For the average person, an AI agent that is vastly cheaper is a much better choice than paying a human lawyer, if that is even possible. The risk of the AI making a mistake is far less than the risk of serious financial distress from paying a human, again, if that is even possible.
And I’m not so sure that with many cases you’re more likely to lose your case with an AI as your lawyer than with the average human lawyer. Maybe there’s some very small chance of hallucinations today, but there’s also a very big chance that the AI is going to find important information and analysis that a human does not know and would not have time to find.
This was bottlenecked by the legal tech vendors until the Claude for Legal announcement just recently. You had to do the legal research inside of Westlaw, Lexis or vLex and they don’t have that capability (unless you automated with browser use, which would be against the TOS and few attorneys knew how to do). But, yes, you’re right - that’s right around the corner.