Understanding The USPTO's New Rules To Create AI Patent Value - Caldwell

As famously explained by Clayton Christensen in The Innovator’s Dilemma, innovation disrupts an industry first by offering a simple and affordable alternative, with initially inferior quality and then, by improving rapidly to compete with and, ultimately, displace incumbents in the market.

Today, widely accessible and inexpensive AI agents, which operate autonomously in making decisions and taking actions, are starting to be used to perform an ever-increasing number of tasks. As these AI agents improve in quality, they will disrupt many industries.

Predicting the impact AI is expected to have, McKinsey estimates that “AI-powered agents and robots could generate about $2.9 trillion in US economic value per year” over the next four years.

Currently, many AI agents produce agentic outputs of inferior quality. However, with technological improvements that focus on specialized market needs and functions, the quality of agentic outputs will improve, increasing agentic AI’s ability to compete with market incumbents.

These technological improvements in AI, which enable better performance of new tasks and disruption of new markets, are now patent-eligible subject matter under new rules at the U.S. Patent and Trademark Office (USPTO).

As an intellectual property (IP) expert, I can share that 2026 presents a rare opportunity to create IP value with patents covering AI inventions.

New Rules

As recently as December 2025, rectifying decades of ambiguity, the USPTO has created new guidance, precedential rules and patent examiner instructions that clarify the circumstances under which the USPTO considers AI inventions patent eligible. Under the USPTO’s new position, an AI invention is patent eligible when it covers a technological improvement.

Accordingly, patent applications filed today are uncommonly well-positioned to be examined under clear and favorable rules. To savvy tech companies, technological improvements enabling AI disruption of major markets will create business value through the expedient filing of U.S. patent applications.

Understanding Patent Examination

Now, during patent examination, patent examiners are required to search for a technological improvement within each AI patent application. Specifically, when searching for a technological improvement, patent examiners must look to two sections of the patent application: the “Specification,” which describes the invention, and the “Claims,” which legally defines the scope of the patent rights.

First, the examiner will search for language in the Specification that describes the invention as an improvement over current technology (and not as an automation of a task currently performed by humans). Then, the examiner will ensure that the Claims are limited to a particular technological field and include non-abstract features producing the purported technological improvement. When the examiner finds a technological improvement described in the Specification and Claims, the AI invention is considered patent-eligible.

To take advantage of the USPTO’s new position, an applicant should write the patent application with these rules in mind. For example, applicants should include an explicit explanation of the technological improvements resulting from the invention in the Specification of the patent application (e.g., the AI is made more efficient, more precise, faster or the like). Applicants should also include in the patent application Claims directly producing the purported technological improvement.

For example, let’s consider the USPTO’s precedential ruling Ex Parte Desjardins. In Desjardins, the patent application’s Specification explained that the invention improved a large language model (LLM) to “effectively learn new tasks in succession whilst protecting knowledge about previous tasks.” Desjardins’ Claims included features that produce this improvement by adjusting “parameters to optimize performance of the machine learning model on the second machine learning task while protecting performance of the machine learning model on the first machine learning task.”

Thinking Ahead

While current USPTO rules are satisfied with a Specification and Claims covering a technological improvement, USPTO rules may change in the future. For instance, the Federal Circuit can overrule, or a new presidential administration can change policies. For fiscal year 2025, the average processing time of a patent application was 31.9 months. For the last three presidential administrations, the USPTO director has changed with each new incoming president. Accordingly, applicants filing a patent application in the second half of 2026 can reasonably expect their application to be examined, at least partially, under a new director—and potentially new rules.

However, by filing an AI patent application under the Track One expedited examination process, applicants can ensure that the application is examined under current USPTO rules. This process requires the applicant to comply with some formal requirements and pay an additional filing fee. By going this expedited route, a patent application will (almost) always reach an allowance or final rejection within one year. So, a Track One expedited patent application filed in 2026 will very likely be examined to completion under the current and favorable USPTO rules.

Capturing Value

AI is poised to disrupt countless markets, creating trillions of dollars in value. Because of this and the USPTO’s current position on AI inventions, companies that prepare and file patent applications on AI inventions today have a remarkable opportunity to capture future technology value.