Practitioner-led, not AI-led
Our approach to AI begins with the practitioner, not the technology. Rather than providing AI with invention disclosure materials or an office action and asking it to develop a proposed strategy for attorney review and approval, our practitioners first evaluate the issues and establish an initial approach based on their own professional judgment. AI is then brought in to help evaluate, challenge, and refine that approach.
This distinction is fundamental. Having an attorney review and approve AI-generated recommendations does not necessarily mean that the attorney has meaningfully participated in their development. Our practitioners are actively engaged in the substantive reasoning throughout the process, using AI to strengthen their own analysis rather than allowing AI-generated recommendations to drive the work.
AI as an intellectual collaborator
We use AI as an intellectual collaborator rather than simply a tool for generating work product. Our practitioners engage AI in substantive, iterative discussions, exploring alternative approaches, questioning assumptions, and evaluating the strengths and weaknesses of different strategies. Just as importantly, we encourage AI to challenge our own reasoning, identify potential weaknesses, and suggest alternatives that we may not have considered.
This differs from workflows in which AI generates work product for a particular task, the practitioner reviews and revises it, and the process moves on to the next task. Instead, we engage AI in a back-and-forth discussion to develop and refine the underlying approach before the work
AI-assisted application drafting
When preparing a patent application, our practitioners review the invention disclosure materials and meet with the inventors to develop a thorough understanding of the invention. We then develop an initial claim strategy and draft at least some of the claims, seeking feedback from in-house counsel or the inventors as appropriate. Throughout this process, we engage AI in substantive discussions to evaluate the proposed claims, explore alternative claim formulations, and identify potential gaps in coverage. The claims are refined through this iterative engagement before we proceed to the remaining portions of the application.
We follow a similar approach when preparing the drawings and specification. Rather than simply asking AI to generate each drawing or each portion of the specification for attorney review and approval in sequence, we establish the initial direction, engage AI in evaluating alternative approaches, and refine the resulting work product through substantive discussion. This active engagement allows us to draw on AI’s capabilities while ensuring that the underlying strategy and content reflect the practitioner’s professional judgment.
AI-assisted office action responses
When preparing a response to an office action, our practitioners begin by reviewing the examiner’s rejections, the cited references, and the prosecution history to develop an initial response strategy, including potential arguments and claim amendments. We then engage AI in substantive discussions to evaluate the proposed approach, identify potential weaknesses, and explore alternative arguments or amendments. Through this iterative engagement, we refine the response strategy before proceeding to preparation of the response itself.
Rather than simply asking AI to propose claim amendments and arguments for attorney review and approval, we use AI to challenge and strengthen our proposed response strategy. This substantive engagement continues through preparation of the response, allowing us to further refine the arguments and amendments as work progresses. Throughout this process, our practitioners remain actively engaged in evaluating the legal and technical issues, ensuring that the resulting response reflects their professional judgment.
Zero data retention, not just no training
Many AI service providers offer assurances that information submitted for inference will not be used for model training. However, this does not necessarily mean that the information is deleted after processing. These providers may retain submitted information for 30 days or longer, for purposes such as monitoring for misuse or enforcing data usage policies.
Our firm goes further by exclusively using AI service providers that guarantee zero data retention (ZDR). Under these arrangements, client information submitted for inference is not retained after processing is complete and the response has been returned. This ensures that confidential client information does not remain stored by the provider, even temporarily after processing, regardless of whether the information would otherwise be excluded from model training.
Geographic restrictions on AI inference
The geographic location where AI inference occurs is not always apparent to the user. AI service providers may distribute inference requests among data centers in different countries—even across continents—to balance workloads and optimize performance. Consequently, even when both the user and the AI service provider are based in the US, confidential information submitted for inference may be processed in another country, potentially without the user’s knowledge.
Our firm addresses this concern by exclusively using AI service providers that provide geographic restrictions on inference. Inference involving client information occurs only in US-based data centers, ensuring that confidential information is not processed outside the US for purposes of AI inference. These restrictions allow us to leverage advanced AI capabilities while still maintaining control over where client information is processed.
Freedom to choose the best AI models
Different AI models have different strengths, and the model best suited for one task may not be the best choice for another. Patent-specific AI applications may limit practitioners to a particular model or a small selection of models, sometimes without disclosing which models are being used. Vendors may also change the underlying models based on commercial considerations, such as the cost of inference, rather than solely on the quality of the resulting work product.
Our approach is to select the AI models best suited to particular tasks, based on our own evaluation of their capabilities and performance. We can use different models for different aspects of patent prosecution and adopt new models as they become available, without waiting for a vendor to support them. This flexibility allows us to more quickly leverage advances in AI technology while keeping decisions about model selection focused on the quality of the services we provide to our clients.
Our workflows, not the vendor's
Patent-specific AI applications often impose their own workflows, requiring practitioners to adapt how they work to the capabilities and limitations of the application. These constraints may force practitioners to use proprietary web-based interfaces rather than their preferred applications, follow prescribed sequences of tasks, or adopt conventions that differ from their established practices. Such restrictions can reduce the flexibility practitioners have in performing their work, potentially compromising the quality of the resulting work product.
Our approach is to integrate AI into our existing workflows, rather than adapting our workflows to accommodate any particular patent-specific AI application. We retain the flexibility to use our preferred software, leverage existing tools and automation capabilities, and tailor our processes to the requirements of individual clients. We can also modify our workflows as new AI capabilities emerge, adopting changes when they improve quality or efficiency rather than because a vendor requires them.
Our expertise stays with the firm
As practitioners gain experience using AI, they may develop specialized instructions, skills, and workflows that improve the quality and consistency of the resulting work product. Some patent-specific AI applications allow practitioners to develop these refinements in their platforms but provide limited or no ability to export them. Over time, the accumulated expertise embedded in these applications can become difficult to transfer elsewhere, creating vendor dependence that grows as the firm’s investment in AI increases. If the firm changes vendors, the loss of this accumulated expertise can adversely affect the quality and consistency of the services provided to clients.
Our approach is to maintain our AI capabilities independently of any particular vendor. The specialized skills, workflows, and other refinements we create remain under our control and are designed to be portable across different AI models and service providers. This allows us to preserve and build upon our accumulated expertise as AI technology evolves, without developing vendor dependence. Most importantly, it supports continuity and consistency in the quality of the services we provide to our clients.
Independence from vendor decisions
Patent-specific AI applications can be costly, reflecting not only the expense of the underlying AI models but also a vendor’s development, marketing, and operating costs. Vendors may change their offerings, discontinue features, or cease operations altogether. A firm that becomes dependent on a particular vendor’s application may face disruptions to its established workflows if it needs to switch vendors, potentially affecting the quality and continuity of services provided to clients.
Our approach avoids dependence on any particular patent-specific AI application vendor. By maintaining our AI capabilities independently, we retain the flexibility to use different AI based on their underlying capabilities and performance. We can adapt to changes in AI availability and technology without abandoning our accumulated expertise or restructuring our workflows around a proprietary application. This independence helps us maintain cost-effective operations and provide consistent, high-quality services to our clients over the long term.