Patent law has always been a peculiar corner of the legal profession, populated by people who hold both law degrees and engineering credentials, who can parse the difference between a dependent claim and an independent claim, and who bill at rates that make even corporate litigators wince. It is also, it turns out, a profession almost perfectly suited for disruption by large language models — and the disruption is already well underway, even if the outside world hasn't noticed.

The work of a patent attorney has traditionally involved three laborious tasks: searching through millions of existing patents to determine whether an invention is truly novel, drafting claims with language precise enough to withstand litigation yet broad enough to be commercially valuable, and monitoring competitors' filings for potential infringement. Each task required years of training and hundreds of billable hours. Each task is now being compressed.

The prior art revolution

Prior art searches — the exhaustive process of scouring databases to find anything that might anticipate a client's invention — used to take experienced practitioners days or weeks. The work was part detective investigation, part keyword archaeology, requiring searchers to anticipate every possible way a concept might have been described across decades of technical literature in multiple languages. AI tools now perform semantic searches that understand conceptual similarity rather than mere keyword matching. A system can identify that a 1987 Japanese patent describing a "rotational force amplification mechanism" might be relevant to a 2026 American filing about "torque multiplication devices" — a connection that would require either bilingual expertise or remarkable luck to catch manually.

The major patent offices have taken notice. The European Patent Office has integrated AI-assisted search tools into examiner workflows. The United States Patent and Trademark Office has piloted similar systems. The private sector has moved faster: firms like Anaqua, PatSnap, and a growing ecosystem of startups now offer AI-powered platforms that have become standard equipment at large intellectual property practices.

Drafting in the uncanny valley

Claim drafting presents a more nuanced picture. The language of patent claims is a specialized dialect — part legal boilerplate, part technical specification, part strategic ambiguity. A well-drafted claim must be narrow enough to be novel but broad enough to capture variations an infringer might attempt. It must satisfy examiners in multiple jurisdictions with different legal standards. It must anticipate litigation that might occur years or decades in the future.

Large language models trained on patent corpora can now generate serviceable first drafts of claims from invention disclosures. They can suggest dependent claims that cover obvious variations. They can flag language that has historically triggered rejections from specific patent offices. What they cannot yet do reliably is exercise the strategic judgment that separates adequate patent protection from bulletproof protection — the intuition about which claim scope will survive both examination and eventual courtroom challenge.

The result is a shift in how junior attorneys spend their time. The associate who once spent weeks drafting claims from scratch now spends days refining AI-generated drafts. The skill being developed is different: less pure drafting ability, more critical editing and strategic oversight. Whether this produces better patent attorneys in the long run remains an open question the profession is only beginning to ask.

The billable hour problem

The economics are uncomfortable. Patent work has historically been billed by the hour, and the hours were substantial. If AI tools compress a forty-hour prior art search into four hours of human review, someone's revenue disappears. Large firms have begun experimenting with fixed-fee arrangements and value-based billing, but the transition is awkward. Clients sophisticated enough to know about AI capabilities are asking pointed questions about why searches still cost what they did in 2019. Clients less aware of the technology are, in effect, subsidizing the learning curve.

Some practitioners see opportunity in the disruption. Patent work that was previously uneconomical — comprehensive freedom-to-operate analyses for startups, continuous monitoring of competitor portfolios for mid-sized companies — becomes viable when AI handles the bulk of the initial analysis. The market for intellectual property services might expand even as the cost per service declines.

Our take

Patent law offers a preview of how AI will reshape knowledge work more broadly: not through dramatic displacement but through quiet compression. The expertise required shifts from execution to oversight, from production to judgment. The practitioners who thrive will be those who understand both what the tools can do and where they fail — who can prompt effectively, evaluate critically, and add the strategic layer that machines cannot yet provide. The billable hour may not survive the transition, but the patent lawyer almost certainly will, albeit in altered form.