Skip to content
ClaudeClaude

Where code meets court: AI at the legal-technical frontier

AI saves lawyers countless hours on research; AI helps developers reason through complex technical systems. Patent law is unique in demanding both simultaneously—researching across millions of documents while comprehending the technical intricacies of novel inventions. The result: a profession undergoing more radical change than any other, and a wealth engineering problems at the frontier of what’s possible with LLMs.

May 22, 202632mWatch on YouTube ↗

CHAPTERS

  1. Why AI fits both software and legal work—and why patent law sits between them

    Ollie Cobb frames two high-value domains for AI: software development and legal work, noting they benefit from different model strengths. He introduces patent law as a hybrid domain requiring both deep technical reasoning and large-scale document analysis.

  2. Patents as a social contract: what a patent grants and why it exists

    The talk explains the core purpose of patents: exchanging public disclosure for a time-limited monopoly. This sets up why drafting must be precise—errors can have long-lived consequences.

  3. The four requirements for patentability: novelty, non-obviousness, utility, disclosure

    Ollie outlines the legal criteria an invention must meet to receive a patent. He emphasizes the “person having ordinary skill in the art” (PHOSITA) standard and the need for sufficient disclosure.

  4. Anatomy of a patent application: claims, spec, drawings, and special formats

    This chapter breaks down the components of a patent application and highlights that claims define legal scope. It also covers support requirements and additional artifacts needed in chemistry and biotech.

  5. Prosecution and the file history: how patents get negotiated over years

    After filing, applicants enter a multi-year back-and-forth with examiners who compare claims to prior art and issue office actions. Responses and amendments build a permanent record that later affects claim interpretation.

  6. Enforcement, infringement, and litigation risk: today’s choices, future consequences

    Once granted, patents can be enforced against products/processes meeting claim elements, and they can be attacked in litigation as invalid. The attorney’s drafting strategy aims to balance breadth with defensibility—often under uncertainty.

  7. Why patent drafting is a uniquely hard AI problem: technical novelty + legal framing

    Ollie summarizes the attorney’s workflow: understand a technical invention, locate true novelty against vast prior art, then encode it in a legally robust document. This highlights why patents need both “reasoning” and “needle-in-haystack” retrieval.

  8. Why generic “delegation” agents struggle in patents (1): outputs can’t be validated like code

    Using software development as a reference, Ollie argues patent work resists the usual agent delegation loop because correctness can’t be quickly tested. Patent decisions are bets against future adversarial actions by examiners, competitors, and litigators.

  9. Why delegation breaks down in patents (2): decisions are tightly entangled across the document

    He contrasts software’s often loosely-coupled micro-decisions with patent drafting’s interdependencies. Changing claim framing can cascade into rewrites of dependent claims, specification support, and drawings.

  10. Additional barriers to delegation: out-of-distribution inventions, hallucination risk, and multimodal artifacts

    Ollie adds further reasons patents are hard to fully delegate: inventions are definitionally novel (OOD), RL-style optimization is less straightforward than coding, and important non-text artifacts (drawings/structures/sequences) require careful representation.

  11. Delegation vs collaboration: the product model Solve targets

    The proposed alternative is collaboration: AI surfaces key judgment calls at the right time and then executes once the attorney decides. This implies different UX and different AI-system design than pure delegation agents.

  12. Three build principles for collaborative legal-tech AI

    Ollie introduces three system principles: make citations fundamental, unify guided workflows with a general agent, and separate alignment from execution by parallelizing preparatory work. These principles aim to increase legibility, control, and efficiency.

  13. Product demo: prior art comparison with clickable citations

    In the drafting module demo, an attorney asks how the invention compares to prior art. The agent reads the disclosure and prior art, produces a comparison (including a table), and provides citations that link directly to supporting passages.

  14. Product demo: guided claim drafting via UI preferences + the same general agent

    The demo shows claim drafting as a repeat workflow with user preferences (claim count, indentation, element label conventions, templates). Even though it’s initiated through a UI, the request is fulfilled by the same agent, preserving reasoning/citation patterns.

  15. Product demo: application review by parallel sub-reviews, then aligned edits

    Ollie demonstrates review as parallel analysis across multiple criteria, producing comments that the attorney can accept/dismiss before triggering edits. This avoids conflicting edits and reduces back-and-forth by separating decision alignment from execution.

  16. Closing summary: choosing domains where collaboration beats delegation

    The talk concludes by emphasizing that valuable AI applications require both strong AI leverage and meaningful application-layer contribution. Patents exemplify a domain where collaboration-centric design—legibility, control, and structured workflows—outperforms pure delegation.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome