Founders File Patents With AI. The Cheapest Draft May Cost the Most

AI can turn messy engineering notes into something that looks like a patent application in minutes. That speed is exactly why startup founders are increasingly drafting and filing patents themselves — and exactly where, according to Michael Drapkin and Nathan Mutter of Landfall IP, an AI-native patent law firm in Boulder, Colorado, the savings can turn into a liability.

The two partners, writing for Crunchbase News, argue the question is no longer whether startups should use AI. It's how much founders can safely do themselves, and when legal expertise becomes essential.

What can AI actually do well?

Plenty, on the preparation side. Drapkin and Mutter say founders can use general-purpose AI tools to:

• organize technical documentation into structured invention disclosures;
• summarize existing patents;
• identify potentially relevant prior art.

Turning raw engineering notes into an organized disclosure can cut preparation costs and let counsel focus on strategy rather than sorting.

But the real opportunity, they argue, goes beyond DIY. Counsel can orchestrate AI across the whole workflow — coordinating invention discovery, drafting, drawings and quality checks. AI can help attorneys explore alternative implementations with inventors, identify disclosure gaps and check whether the claims have support in the specification. That combination produces more comprehensive applications with substantially greater efficiency.

"It is fundamentally different from generating an application with a chatbot and asking a lawyer to review it afterward," the authors write.

Where does DIY filing break down?

The problem, per Drapkin and Mutter, is that AI's capabilities are jagged: it performs some tasks remarkably well while missing the legal and strategic issues that determine a patent's commercial value. An application can look sophisticated while providing little meaningful protection.

A patent's claims define its legal boundaries — what you can stop competitors from doing. Claim-scope mistakes can stay hidden until an investor, acquirer or competitor scrutinizes the company's intellectual property.

The authors single out three specific failure modes:

• Incomplete provisional applications. A provisional's early filing date is only as useful as its disclosure. Later claims receive that date only if the provisional adequately describes and enables the claimed invention. A DIY draft may omit technical details, alternatives and variations essential to future claims — and adding those details later cannot repair the original disclosure retroactively.
• Premature disclosure. Pitching investors, demoing products or announcing innovations before filing can jeopardize patent rights, particularly internationally, and can destroy trade secret protection. Feeding confidential invention details to AI tools without checking their data practices adds further risk.
• Unclear ownership. Paying an employee or contractor doesn't substitute for signed invention assignments. The authors advise securing those agreements when people join — not waiting for diligence to expose gaps — and verifying obligations to prior employers or universities. Ownership gaps can derail fundraising or acquisitions.

Their bottom line: the cheapest filing option is not necessarily the most economical. A poorly prepared application creates false confidence while leaving valuable innovations exposed.

When does counsel matter?

Lawyers earn their fee on judgment calls, the authors argue: deciding what to patent versus keep as a trade secret, prioritizing filings within budget, and assessing deadlines before disclosure or statutory limits narrow the options. An internal process competitors cannot observe or reverse engineer may be better protected as a trade secret than by any filing.

Claim scope requires particular judgment. A fraud-detection claim tied to one algorithm leaves competitors free to deliver the same functionality with another. Overly broad claims, meanwhile, may overlap with existing technology. The valuable claims target what competitors must do to compete — ideally, a single identifiable competitor can infringe them, and that infringement can be detected and proven. The specification, in turn, must describe alternatives and variations that support broader claims later without unnecessarily limiting their scope.

Why do investors look past 'patent pending'?

Filing volume says little about commercial value, the authors write. Investors and acquirers look beyond the count of pending applications and ask three questions:

• Do you own it? They want signed assignments from relevant founders, employees and contractors, with no competing claim from a prior employer or university.
• Does it protect your moat? Claims should cover what the business depends on — not a feature it will abandon next year — and the application must contain enough detail to support them.
• Can you enforce it? Claims should be hard to design around, and infringement should be realistically detectable and provable.

Who are the authors?

Drapkin, Landfall IP's founding partner, previously built one of the largest electronics and computer patent prosecution practices in the U.S. at an Am Law 100 firm and was named U.S. Patent Practitioner of the Year by Managing IP in 2022. Mutter, the firm's managing partner and founder, has served as top outside counsel at Oura, Coinbase, Rubrik, Salesforce and other startups and public companies, and pioneered automation systems across legal services.

Their advice to founders shopping for counsel: ask how a firm coordinates its AI tools and legal expertise — not simply whether it uses AI. The answer should explain how that process improves both efficiency and substance.

AI's greatest value, they conclude, is not generating more applications at lower cost but making meaningful protection more accessible without sacrificing legal judgment. The most expensive patent mistake, they warn, may be discovering too late that a DIY application never adequately protected the innovation on which the business was built.