This approach does not remove the attorney from the process. Instead, it changes where their involvement is concentrated, moving some of the work away from sentence construction and toward reviewing the strategy, scope, and substance of the application.
Your filing targets have increased this year, yet your team’s size remains unchanged. Meanwhile, the tools provided to aid you are largely just existing software with a chat window added: helpful for creating a summary of an office action, but unhelpful for the 40 to 80 hours between submitting an invention idea and actually filing an application. This gap is why the patent industry is starting to divide into two groups: AI-based tools, where intelligence has been added as a feature to existing software, and AI-native platforms, where the intelligence itself is the software. The difference is fundamental, not superficial, and determines exactly what task you can outsource from your workload.
The removal test
There is an easy way to tell these products apart. Think about taking away the AI capabilities from any tool your group uses and see how well it performs. If the product still functions much like it did back in 2019, with similar screens and workflows, although perhaps missing a “summarize” button, it is AI-based. The AI was incorporated later. On the other hand, if the product falls apart without the AI, it is AI-native. The AI has always been a key part of it, and the review controls, audit trail, and entire data model were developed around it. Investors providing the funds for this changeover also apply the same test.
The difference might seem theoretical if our field were not constantly evolving. But we’re not standing still. Research from Clarivate TM found that AI adoption within IP professions increased from 57% in 2023 to 85% in 2025, and its analysts now refer to non-adoption, rather than adoption, as the position requiring justification. Capital is flowing in too: in the span of about five months from late 2025 to early 2026, three patent-AI platforms completed funding rounds totaling $100 million, with long-established companies like Thomson Reuters appearing on the list of investors.
Even the patent office is getting involved. USPTO examiners are conducting AI-assisted prior art searches using the Similarity Search function within the office’s own tools, something the USPTO itself documented in April 2024. Considering that first office action pendency reached 22.6 months during FY2025 and there is an ever-growing backlog of unexamined applications that peaked above 837,000, the examiner reviewing your application will have machine-ranked art in front of them. An application written without equivalent research depth arrives at that meeting severely underprepared.
What AI-native means in patent practice
Architecture really counts here, and that’s where Esgenix resides.
The first wave of patent drafting software worked one prompt at a time: ask for a paragraph, get a paragraph, then stitch it all together yourself into an application. Instead, an AI-native platform revolves around “agents.” An agent is essentially a program set with a specific objective and allowed to execute multiple steps to reach it, perform searches, create sections, and verify its output against the disclosed information, rather than simply responding to a query and halting. “Agentic” drafting refers to several of these agents tackling the issue sequentially, just like a well-coordinated group would, under the watchful eye of the attorney reviewing and validating regularly.
Why the native architecture matters can be seen clearly through search functionality. Esgenix runs an AI patent search that matches the meaning of the invention rather than relying only on keywords, across 220 million documents from 108 jurisdictions. In an AI-driven workflow, this type of search may operate as a separate tool, with its results reviewed before drafting begins. In Esgenix, the search results are available to the drafting agents during the same session. This allows the claim set to be considered alongside relevant prior art from the beginning. The specification can account for areas where the prior art is dense, while the figures can focus on features that distinguish the invention. In this workflow, the search is not treated as an isolated step. Its results can inform the claim set, specification, and figures while drafting is underway.
The same integration can take place before drafting begins. When the invention disclosure is provided, the platform identifies potential gaps, such as unresolved steps, undefined terms, or results that lack supporting explanations, and provides suggestions for addressing them while the inventor is still involved. During claim-strategy planning, the scope of the claim set can also be adjusted using a slider, with supporting information in the specification updated based on those changes. The approach extends to the figures as well. Rather than requiring an existing sketch, photograph, or CAD file as a starting point, Esgenix can work from the written specification to determine what figures are needed and generate them, while cross-referencing the reference numbers in the drawings with the corresponding text.
The attorney does not leave the loop
Nothing in this process goes unchecked, and it shouldn’t. The right comparison isn’t a machine operating autonomously; rather, it’s assigning tasks to an experienced senior associate with prearranged touchpoints. Within Esgenix, these touchpoints take the form of five specifically defined gates, and the matter won’t progress past any of them without your direct involvement. Initially, at intake, you verify the invention, address technical gaps, and establish the relevant jurisdictions. Next, during strategy, you determine the breadth of the claims using the scope slider and select the fallback positions. At the claims stage, you can approve or revise the proposed claim set. The process does not proceed automatically. For the specification and figures, you review the complete draft, with provenance displayed alongside each claim component, showing where each element originated in the invention disclosure. At sign-off, a human being provides the final approval.
This approach does not remove the attorney from the process. Instead, it changes where their involvement is concentrated, moving some of the work away from sentence construction and toward reviewing the strategy, scope, and substance of the application.
These review points are particularly relevant given the documented risks associated with unchecked AI use. By mid-2026, there were more than 1,500 documented court cases involving AI-generated citations in a publicly accessible database. The USPTO’s April 2024 guidance also makes clear that, while there is no general requirement to disclose the use of AI in drafting, relying solely on an AI tool’s accuracy does not satisfy the practitioner’s duty to make a reasonable inquiry.
A workflow designed for patent-related tasks therefore needs a structured way to support human review. Provenance provides one part of that process. Another is a quality-control step known as an LLM-as-a-judge, in which a separate model evaluates the document against defined standards and can flag areas for revision before the attorney completes the review.
Two operating models, side by side
The architectural difference produces a commercial one. AI-based tools are still sold like the underlying software, by seat, every year, based on headcount. Esgenix prices by matter delivered, across different tiers, so costs follow filings more closely than team size. That’s a fundamentally different approach for a large corporate IP group. It is not about getting a better price; it is about structuring the budget based on actual output.
One thing we need to be careful about when describing this change is that some vendors in our market say they are alternatives to patent law firms in their own marketing materials. Esgenix doesn’t say that either. The system natively creates drafts for US, EP, and Indian practice, applies a preset house style to every agent per client, and can perform white space analysis and freedom to operate checks within the same drafting session. However, it doesn’t file anything, make any decisions, or provide final approval. All those tasks remain yours, by design.
That’s really what differentiates an AI-native patent workflow from just another AI function integrated into current software. The real question isn’t whether a tool uses AI. Rather, it’s whether the whole workflow was developed around what AI can generate while retaining the key decisions for the attorney.
