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 ↗

WHAT IT’S REALLY ABOUT

Why patent drafting needs collaborative AI, not delegated agents alone

Patent law sits at the intersection of deep technical reasoning and large-scale document search, making it a uniquely high-leverage domain for AI assistance.
Unlike software, patent work is hard to validate with tests and is shaped by adversarial, long-term uncertainty, so “delegate then verify” breaks down.
Patent drafting decisions are tightly entangled across claims, specification, drawings, and later prosecution history, requiring sequential human judgment rather than a single autonomous run.
Solve Intelligence proposes a collaboration-first product model grounded in (1) first-class citations, (2) guided UIs that still route through a general agent, and (3) parallelizing analysis while sequencing user alignment and final execution.
A product demo illustrates prior-art comparison with source-linked citations, claim drafting via a structured interface feeding an agent, and multi-criteria application review using parallel sub-reviews followed by aligned edits.

IDEAS WORTH REMEMBERING

5 ideas

Patent work benefits from AI for different reasons than coding does.

Software leverage comes from abstract technical reasoning plus fast validation, while legal/patent leverage comes from filtering massive corpora and extracting what matters under formal rules and future disputes.

In patents, you can’t “unit test” correctness, so autonomy is riskier.

A patent’s quality is judged by future examiner rejections, competitor design-arounds, and litigation invalidity attacks, making output evaluation slow, uncertain, and high-stakes.

Entanglement across claims, spec, and drawings makes late-stage review insufficient.

Changing claim scope or terminology can force cascading updates throughout dependent claims, supporting passages, and figures, so attorney judgment must be applied continuously as the document forms.

Out-of-distribution inventions and costly hallucinations push systems toward evidence-backed assistance.

Because each invention is “new” by definition and hallucinations can hide in technical/legal nuance, the system should emphasize verifiable sources and careful reasoning rather than improvisation.

Treat citations as infrastructure, not a cosmetic add-on.

To create a real audit trail, every piece of information shown to models (PDFs, disclosures, pulled prior art) must be represented in a citeable way, and agents/subagents must preserve attribution end-to-end.

WORDS WORTH SAVING

5 quotes

At their core, a patent is a social contract between an inventor and society.

— Ollie Cobb

By contrast, you can't run a patent unfortunately, and its correctness is really a function of events that haven't happened yet.

— Ollie Cobb

So really the decisions you're making are like bets against that adversarial future, and it's not like they're really right or wrong, but they trade sort of one type of risk for another based on the sort of risk appetites of those who will bear the consequences.

— Ollie Cobb

But together, the attorney's judgment can't be deferred to some final review pass, and it can't be sort of concentrated up front, but instead needs to be sort of imparted sequentially as the patent comes together.

— Ollie Cobb

And so really, what we need here is something more akin to a model of collaboration rather than merely delegation.

— Ollie Cobb

Patents as a social contract and core patentability criteriaPatent application structure: claims, spec, drawings, abstractProsecution workflow and file history consequencesWhy delegation works for coding but not patentsValidation difficulty and adversarial future riskEntangled drafting decisions across document componentsCollaboration-first AI principles: citations, workflow-to-agent translation, parallel alignment/executionProduct demo: prior art comparison, claim drafting, application review

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

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