I am working on a simple question with surprisingly large implications:
What would legal services look like if we designed the work from first principles — around the tasks that need to be completed rather than the people who traditionally perform them?
We started with contract law and mapped the complete lifecycle:
Find the work → Produce the work → Manage the result → Identify the next need.
That gave us 45 distinct tasks, ranging from identifying a potential legal need and establishing facts to legal research, analysis, drafting, negotiation, monitoring obligations and detecting the next emerging issue.
Then we asked a different question for every task:
Does a human actually need to perform this — or does a human need to remain responsible for what happens?
Those are not the same thing.
Our first architecture exercise suggests that a substantial part of legal production could eventually run through AI, agents and deterministic orchestration, while humans increasingly concentrate on mandate, judgment, relationships, trade-offs, decisions and accountability.
We are now turning that observation into an architecture for an AI-native legal operating platform built around:
Human → Agent → Agent → Human.
The problem I am working on is therefore no longer simply:
How can AI help lawyers?
It is:
How should legal work itself be designed when machines can perform much of the production, while humans retain authority exactly where human judgment matters?