Invention Analysis Workflow

Question

Can a structured multi-agent workflow produce a useful preliminary invention analysis?

Objective

Test whether breaking invention analysis into intermediate, inspectable AI work products — rather than a single conversational answer — produces output an attorney can usefully review and revise.

Background

Invention analysis typically requires understanding a disclosure, identifying the inventive contribution, and assessing it against context only an attorney can supply. A single conversational AI response does not naturally provide the structure that analysis requires — see The Problem on the Workflows page for the Laboratory's broader framing of this gap.

Hypothesis

A structured pipeline — disclosure → invention analysis → patentability assessment → fallback positions → attorney review — will surface a usable preliminary analysis, though most likely one that requires meaningful attorney revision rather than a finished work product.

System

Conceptually, this experiment follows the Invention Analysis workflow described on the Workflows page's Example Workflows: a disclosure is analyzed by AI, assessed against patentability considerations, and reviewed by an attorney. The underlying implementation has not yet been published in detail.

Method

TBD — the experiment's method will be documented once it has been run.

Results

TBD — no results have been published yet.

Evaluation

TBD — evaluation criteria will be published alongside results.

Failure Modes

TBD — failure modes will be documented as they are observed.

Human Review

TBD — this section will describe where attorney review proved necessary once the experiment has been run.

Lessons

TBD — lessons will be published once there is something to report.

Next Experiment

TBD.