Harvey Wants Its AI to Stop Starting From Scratch
Harvey has launched Harvey II, introducing a new architecture built around matter context, persistent memory, agents, and legal-specific intelligence.
Harvey II uses Spaces to preserve the documents, parties, tasks, permissions, history, and other context associated with a matter or project.
Agents operating inside a Space can inherit that context rather than requiring lawyers to repeatedly upload documents and explain the matter.
The platform also introduces persistent user memory.
Harvey says the system can learn preferences such as how an individual lawyer structures summaries, handles citations, or writes.
Memory can follow the user across Harvey, Word, and Outlook, and users can inspect, modify, or disable what the system remembers.
Harvey says this memory is not used to train its models.
Harvey II also connects this context layer with Harvey Tenet, its newly post-trained legal model.
Why It Matters
Most generative AI still behaves as though every interaction is the first conversation.
That is fundamentally different from how professional work operates.
Legal work carries context.
A lawyer knows:
- what happened earlier in the matter
- what the client prefers
- which positions have already been taken
- how the partner likes work structured
- what precedent the team uses
- what the relevant ethical restrictions are
Repeatedly reconstructing that context creates friction and limits how much work can realistically be delegated to AI.
Persistent context begins to change that.
Instead of an AI tool answering isolated questions, the system starts becoming aware of the matter, the workflow, and the individual lawyer.
The Crestpoint Take
Memory could become one of the most important layers in professional AI.
There are really several types of intelligence beginning to converge:
General Intelligence
What the underlying model knows.
Domain Intelligence
What a legally post-trained model understands about legal work.
Institutional Intelligence
What the firm knows from its documents, precedents, workflows, and experience.
Matter Intelligence
What is relevant to this specific client or engagement.
Personal Intelligence
How this particular professional works.
Harvey II is beginning to combine those layers.
That has enormous potential.
It also creates new governance questions.
What should the system remember? Who owns that memory? Which information belongs to an individual versus a team? What happens when a lawyer changes matters, practices, clients, or firms? How should ethical walls apply to learned preferences and contextual information? How should institutional knowledge be separated from personal preference?
Those questions will become increasingly important as AI moves from stateless assistants toward persistent professional agents.
What to Watch
The most important development may come when individual memory evolves into team and organizational memory.
If firms can safely encode preferred drafting approaches, negotiation positions, workflows, playbooks, and accumulated expertise into persistent AI context, AI may begin to function as a new layer of knowledge infrastructure.
At that point, the conversation stops being primarily about productivity.
It becomes about how professional organizations capture, preserve, and operationalize expertise.