The hard problems are still ahead
The hardest problems in AI begin long before there is an answer to find.
Truly understanding the matter requires intelligence to reason not only about the world, but about its own understanding of it: why it believes what it believes, where the limits of that understanding lie, and what it still needs to learn. It means closing the loop between knowledge, reasoning, and learning, and discovering not only the answer, but questions that no one knew to ask.
Legal work brings these problems together at unusual depth and scale. They are not uniquely legal problems. They are problems at the core of general intelligence.
Legal AI is an AI research problem
These questions long predate today's generation of models. Legal AI does not sit apart from the broader problems of intelligence. It concentrates many of them in one domain.
New capabilities have changed what is possible. They have not made the underlying problems disappear.
The frontier is Reasoning at Scale
Today's AI systems tend to do one of two things well: reason deeply over a relatively small amount of information, or skim much larger amounts of information with limited depth.
The frontier is… doing both.
As the information space grows by orders of magnitude, reasoning does not simply scale with it. Progress requires returning to first principles: how knowledge is represented, how reasoning is carried out, and how the two interact at scale.
At that scale, intelligence can no longer be contained in the model. It emerges from the system as a whole: models, algorithms, and infrastructure working together.
The same is true of the people building it: research and engineering can no longer be separated.
Rethinking work
Much of legal work today reflects the limits of the technologies that shaped it. Tasks became workflows, workflows became tools, and boundaries hardened around what software could and could not do.
As those limits change, those boundaries become open questions.
The opportunity is not simply to automate the work as it exists, but to reconsider how the work should be done in the first place.
That same principle applies to us. If we are rebuilding legal work from first principles, we should be willing to rebuild how we create the technology behind it.
For us, research and engineering are not two steps in the process. They are one.
We are lawyers, scientists, and engineers. But at heart, we are all builders.
Better legal AI begins with AI research.
