Joel Hron

you know, haven't worked with many for the last several years.

Like, you know, you go to law school to learn like what the patterns of good research look like and what sort of the, you know, way to formulate an argument is.

And there's precedent for how to formulate good arguments or how to like argue against arguments.

And so there is a structure to how good label research should be done.

And then the second part, and so that's all on the like AI reasoning side of the equation.

The other side of the equation is like, okay, what tools and content and information does this agent have available to it to go do that research?

And as Thomson Reuters, this is sort of one of the things that we've been exceptional at for many, many years.

We have probably the world's largest and most respected research.

repository of content for this type of thing.

But we haven't just sort of like put it on top of a bunch of raw content.

There are aspects of our legacy software tools like Westlaw, for instance, that really are tools that human lawyers have used for decades to do good legal research.

An example of that will be something like Keysight.

Keysight is like a feature of Westlaw that basically annotates whether a new case conflicts or overrules a prior case along a particular topic.

And so this is a tool that a user in Westlaw would use to say, oh, I'm looking at this case.

Like, I think it supports my argument, but maybe it's no longer good law because this new case came out.

I should go follow that Keysight flag and look at that new case perhaps.

And we have trained our agent to use Keysight in the same way, for instance, that a human lawyer would use it.

And so the agent now is able to see these breadcrumbs and follow these breadcrumb trails to do better and more comprehensive research than it would have done otherwise.

And I think that combination of both injecting the AI with good sort of legal grounding in how this research is to be done and also arming it with world class tools that are really kind of tuned for how the agent needs to interact with them.

Is a key ingredient of of building, you know, good agentic systems and in particular building good deep research systems.