Why Patent Practice Needs Purpose-Built AI: Reflections from Code with Claude

Key insights

• Patent work sits at the intersection of deep technical reasoning and large-scale document analysis, two areas where Claude excels
• Domain-specific software adds value beyond what general-purpose AI tools can offer
• The most consequential design choice for AI in patent practice is keeping the attorney in control

Why patent work is a good fit for modern AI

The talk opened on the point that patent work sits at the intersection of two capabilities modern AI handles well: technical reasoning over complex subject matter, and identification of relevant material across large document sets. 

Software development tends to draw mainly on the first. Much of legal practice draws mainly on the second. Patent work draws on both at once. Drafting a claim requires reasoning about what is technically novel in an invention while keeping the constraining prior art in view. Responding to an office action requires engaging carefully with the examiner's technical reasoning while working through the cited references to find the strongest counterargument. 

In each case depth and breadth have to be brought to bear on the same problem at the same time, and the fact that models have become capable of both is much of what allows AI to be so valuable for patent work

Domain-specific software is where the value compounds

A second theme was that the value of an AI application does not come from the model alone. It comes from how the application is designed around the model: the interfaces, the workflows, the way information is sourced, structured, and cited. 

For patent work specifically, this means tooling that respects how attorneys actually work. The way prior art is surfaced when an attorney is reasoning about claim scope is one example. The way citations are handled, so that an attorney can verify each part of an output against the underlying source, is another. So is how non-textual material, figures, chemical structures, biological sequences, is represented to the model and to the user. These are the kinds of details that purpose-built software gets right and general tools do not.

Building closely with Anthropic

Solve Intelligence has been powered by Claude since the platform's earliest days, and our partnership with Anthropic has continued to expand, most recently through the launch of our MCP server inside Claude. This MCP connection means that users who have both a Solve Intelligence subscription and a Claude subscription can use some basic Solve features directly within Claude, so long as the connection is enabled. Speaking at Code with Claude reflects the depth of that collaboration, and the shared view that the future of applied AI in regulated fields will be defined by the application layer as much as the model.

A practical implication for firms and in-house teams

For practitioners weighing where AI fits in their workflow, the takeaway from Code with Claude is a practical one. The model under the hood is converging across serious applications, which means the differences that matter in daily use are coming from elsewhere, from how the software handles citations, how it surfaces prior art, how it represents figures and structures, and how it keeps the attorney in control of consequential decisions. These are the kinds of details that purpose-built tools get right and that AI features added to general-purpose software tend to miss. The shape of the tool matters as much as the model behind it.