Purpose-Built for Patent Prosecution: Why ARTY is Not a General-Purpose AI Tool 

Purpose-Built for Patent Prosecution: Why ARTY is Not a General-Purpose AI Tool

Most AI tools used in legal work were not designed for legal work. They were designed for general documents and adapted to legal contexts. ARTY was designed the other way around. This is the third article in a four-part series on ARTY. 

There is a category of AI tool in the legal market that does many things passably and a few things well. It can draft contracts and patent applications and memos. It can summarize transcripts. It can answer general legal questions with varying accuracy. It is general-purpose, which is a feature in some contexts and a limitation in others. 

Patent prosecution is one of the contexts where general-purpose tools start to struggle. The work is highly structured, the stakes are high, and the conventions are specific. Antecedent basis is not a stylistic preference; it is a §112 requirement. Claim dependency is not document hygiene; it is a question of scope. The right model to use on a §101 eligibility analysis is not the right model to use on a routine declaration form. A general-purpose AI tool cannot know that. It treats everything as a document and reaches for whichever model happens to be its default. 

ARTY was designed with the specific shape of patent prosecution in mind. Most of what follows is the consequence of that design choice. 

Built Around the Firm’s Docket Structure 

ARTY’s data model mirrors the way SLW manages prosecution work in our docketing systems. Matter numbers, application families, prosecution timelines, and the relationships between continuations and parent applications are first-class concepts in ARTY, not metadata bolted onto a generic document store. When an attorney opens a matter in ARTY, the tool understands what kind of object it is looking at, and where that object sits in a larger family. 

Because ARTY is built on standard, enterprise-grade infrastructure on AWS, the platform can integrate cleanly with the firm’s other operational systems. The matter that lives in our docketing system is the same matter that ARTY references when it retrieves prior art, prosecution history, or a claim set. There is no manual mapping step, and no parallel data structure to keep in sync. 

“Generic AI document tools are built for the average law firm’s average workflow. Patent prosecution at scale is not an average workflow. The data structures, the file formats, the relationships between matters, the conventions of claim drafting and office action responses, none of it is generic. The tools we build need to reflect that.”  — Suneel Arora, SLW 

Patent-Specific AI Routing 

Different prosecution tasks benefit from different AI models. A quick syntactic check on antecedent basis runs faster and at lower cost on a smaller model. A complex §101 eligibility analysis benefits from deeper reasoning. A multi-claim restriction analysis with examiner history may benefit from yet another tradeoff. 

ARTY routes between foundation models from Anthropic, OpenAI, and Google based on task type and complexity, with most inference flowing through AWS Bedrock. The routing logic is not generic document-complexity heuristics. It is shaped by patent-specific signals such as claim complexity, office action type, and prior art volume. 

The result is a tool that uses the right model for the right task, rather than burning expensive inference time on routine work or sending complex analyses to models that cannot handle them. 

Jurisdiction-Aware Patent Data 

Prosecution work is not bounded by a single jurisdiction. A U.S. patent family often has counterparts in Europe, Asia, and elsewhere, and the data feeding any prosecution decision needs to reflect that reality. 

ARTY draws directly on USPTO data sources, including Patent Center and the bulk data programs, for U.S. prosecution histories, application status, patent documents, and examiner data. For European matters, ARTY integrates with the European Patent Office’s Open Patent Services and Espacenet. 

The platform’s data layer also draws on OTTO Hub, the data platform from SLW’s strategic partner Black Hills AI (BHAI). OTTO Hub provides structured, continuously maintained access to a broad range of patent data: matter audits, file histories, bibliographic and continuity data, prosecution references, and on-demand reference retrieval across U.S. and international jurisdictions. Treating that as a foundational third-party data source rather than something we would attempt to assemble ourselves means ARTY’s drafting agents are reasoning over current, authoritative record data, not over scraped or stale sources. 

“World-class third-party data sources are what make truly custom tools possible. OTTO Hub, through our partnership with Black Hills AI, gives ARTY structured patent data with a depth and reliability that would be impractical for any single firm to build and maintain at the same standard. Layered with our firm-specific data and prosecution patterns, that foundation is what lets us build workflows tuned to how SLW actually practices rather than the average use case. Custom tools depend on tight integration with the data underneath them, not on any single model or vendor on top.”  — Steve Lundberg, SLW 

Multi-Model Intelligence, by Choice 

One technical decision worth flagging: ARTY is not single-vendor. It routes between models from Anthropic, OpenAI, and Google, and the architecture is built so that adding new providers or models is a configuration change, not a redesign. The decision was deliberate. The AI provider landscape is shifting quickly, and committing to a single model family forecloses options the platform may need to take advantage of later. The model layer is the part of the stack most likely to change. The data foundation underneath, and the patent-specific logic on top, is what compounds. 

Existing Systems, Not Silos 

The platform integrates with the firm’s existing systems rather than sitting alongside them as a separate silo. The matter in ARTY is the same matter in our docketing system. The document that ARTY drafts is exportable to the formats and workflows our prosecution teams already use. The validation reports ARTY produces feed into the same review workflows our attorneys already follow. 

This integration matters less for individual matters and more for portfolio-level work. When a client has hundreds or thousands of patents in flight, the value of having the firm’s tools speak to one another is substantial. ARTY is a part of that integrated picture, not a separate system attorneys log into for a subset of tasks. 

What This Means for Clients 

Clients sometimes ask us why SLW chose to build ARTY rather than license one of the AI patent tools on the market. The honest answer is that the available tools, while increasingly capable, are not designed for the specific shape of how we practice. The data structures, the docket conventions, the integration with our other firm systems, and the model-routing choices that make sense at SLW’s scale are not the choices a general-purpose tool would make. 

Building ARTY around the shape of our practice means the platform compounds in value over time. Each iteration adds capability that maps directly to the firm’s work, rather than capability we then have to adapt to the firm’s work. For clients, the result is a tool whose investments accumulate in directions that benefit their matters. 

The next and final article in this series will turn to what ARTY does not do, which is to say, the work that stays human, and how the platform is designed to expand attorney attention rather than substitute for it.