[IP Practice Online] Transforming corporate IP with AI agents — From business assistant to strategic partner (Kosuke Kuriyama)
Updated: 4 hours ago
Article information
Categories: Patents
Topics: AI and intellectual property / IP management
Published on this website: 2026-09-29
[IP Practice Online] Transforming corporate IP with AI agents — From business assistant to strategic partner (Kosuke Kuriyama)
[Summary translation to fit the character limit]
IP departments should redesign operations and organization around AI that obtains materials, plans and executes multiple processes. The aim is to assign routine filing, prosecution and searches to AI, freeing humans for case prioritization, claim design, business alignment and IP strategy.
1. Chat AI answers individual questions; refined prompts improve answers; agents plan and use files, databases, APIs and searches autonomously; multiple agents can cooperate while humans set vision and evaluate. MIXI's IP office mainly uses Claude Code in VS Code.
2. In the demonstration, a PDF of dispatched documents lets AI identify the case, retrieve office actions, references, specifications and claims, analyze refusal grounds, propose responses and draft arguments and amendments. For allowed applications, it collects families and claims, evaluates the patent and proposes divisionals and claims.
3. Human involvement varies with importance: ordinary prosecution is broadly assigned to AI; humans decide divisional need and policy and closely handle important claims and filing strategy. Multiple proposals, including grant-prioritizing and challenging options, counter easy narrowing amendments. Important cases receive special flags/routes (around 30 minutes).
4. Filing has human review between stages. FTO work proceeds from planning materials and queries through thousands of patents, multistage screening, detailed analysis and reports. AI queries may still be inferior to specialist searches; broad retrieval followed by AI noise removal remains experimental.
5. Articulate decision criteria, reasoning, formats and logic; define task-specific boundaries for AI reliance and human review; continually verify inputs, processing and outputs and improve prompts/scripts. MIXI manages these in GitHub: about 18,000 filing-related lines and 9,000 prosecution lines.
6. Humans should direct multiple AI operations. The stated goal is five people producing the output of 30. Surplus time goes to IP analysis, business alignment, portfolio design and management proposals, not headcount reduction.
7. External specialists remain necessary. Companies may internalize routine documents using AI, increasing the value of external patent attorneys who make tacit knowledge, criteria and reasoning explicit for AI. The goal is an IP department that presents strategy to management and the business, beyond administrative efficiency.
Originally posted on LinkedIn: https://www.linkedin.com/feed/update/urn:li:activity:7510140032020602880/
Comments