Business people, mostly outside of AI (lately, a lot of lawyers), keep talking about how easy it is to make AI handle contracts.
Thankfully, my team knows better.
I think the confusion works like this: summarization is easy. Tens of millions of contracts are in public datasets.
So... can an LLM bluff being a lawyer?
No. Not quite.
That’s because summarization, even really good summarization, isn’t all that useful. A contract only has value in the problem space if you can validate against it. Meaning: what situations does it cover, what outcomes does it drive, and how do you check invoicing, billing, or performance against those terms?
The answer is to adjust your mind model.
Rather than a static document, imagine a contract as a living control center for business relationships and outcomes. It lays out, hopefully clearly, what the rules of the road are, and then it builds the expressway that you do business on.
But I get why it’s hard to appreciate until you’ve actually tried to build a system that can handle 200+ pages of verbose, interconnected prose.
This is one of those long horizon problems you keep hearing about.
Doing it correctly means understanding the layers of logic and operational norms for any given industry, and how the specifics of any given agreement play out over time.
It’s all doable.
And in my opinion, it’s some of the most meaningful, potentially life saving work in AI right now.
But easy is not the word I would use to describe it.