“Just when I thought I was out, they pull me back in.” That’s Michael Corleone in The Godfather Part III, and lately it’s been my line too. About five years ago I mostly stopped writing this blog. I was in the trenches, doing the work rather than writing about it. What pulled me back in wasn’t the family business. It was AI. It’s turning legal tech upside down, and I couldn’t sit this one out.
Two Surprises
Coming back, I found two surprises. The first is that this stuff is fascinating. I’ve spent my career with document assembly and document management, and I didn’t expect to be this excited about a new tool. The second surprise is that I was able to make time to do it. For someone who’s been running a major migration every week, that’s no small thing. I’ll come back to where that time came from.
A Replacement? No. An Enhancement.
AI has been sold as a replacement for document assembly. After about a year of research, examination and testing, my answer is categorically no. But that doesn’t mean it has no major role. It enhances the document assembly process.
Here’s where it helps. Building a complex drafting system means balancing two kinds of language: the terms only a specialist estate planning lawyer would understand, and the terms a smart doctor or businessman would. AI can bridge that. It can take unstructured information from a client, from a friendly interview or a request for structures, and turn it into the answers a document assembly system needs. That covers hard facts like names and addresses, and needs like “my estate is getting too large and taxable.” Tools now on the way for HotDocs Advance will let you upload documents from inside an interview, extract the answers, then flag or improve them.
AI can also review a firm’s past precedents to see what changes from deal to deal, and it can check a document for consistency: cross-references, numbering, anything out of place.
Deterministic versus Interpretive
The real difference is simple. Document assembly is deterministic; AI is interpretive. For a lawyer, that matters. AI isn’t predictable, so its work needs review. A deterministic system takes fixed inputs and gives completely predictable outputs.
So the sweet spot is using both. AI helps you pull out the values that fill in the blanks, and it can take a pile of scattered facts and suggest a recommendation. Then the document assembly system takes those clarified, fixed inputs and produces a document that needs no further legal review.
There’s one more twist. The hard part of building a good document assembly system has always been knowing, in advance, every branch and alternative a firm uses when it drafts. Most firms work from a form they edit and retype. Now I use AI as an interpretive tool to build the system itself. We mine the actual drafts, see what’s varying, ask the lawyers why, and build more branching rules in. Then we do it again.
Five Years in the Trenches
So what have I been doing for the past five years? Not sleeping much. My migration work went from a few groups of lawyers a month to major law firms, a major migration every single week, and moving 200 very busy attorneys in a week is a different world from moving ten. Alongside that, I’ve been taking complex multi-template HotDocs systems to the cloud and into client-facing applications. AI has yet to earn its keep in migrations, but that’s what kept me busy.
Where the Time Came From
So where did the time come from? It’s not that AI agents are off doing my bidding. They’re running the scriptlets, processes and steps I would have done myself, or trained another engineer to do. That frees me to design the process and manage the agents, rather than just do the work. It’s given me downtime to think about bigger issues, and to write again. Just when I thought I was out, AI pulled me back in.