PatSnapが示したAIエージェント型特許調査・技術探索

先日、PatSnapの佐藤氏による「手戻りなき最速の開発サイクルへ」と題したセミナーを聴講しました。テーマは脱レアアースモーター開発でしたが、本質はAIエージェントによる研究開発・知財業務の変革にありました。
講演では、脱レアアースというゴールに対して複数の技術ルートが存在し、どのルートを選ぶかによって開発期間やコスト、成功確率が大きく変わることが説明されました。多くの開発プロジェクトでは、初期調査が不十分なまま設計や試作に進み、後になって特許障害や技術的課題が見つかり、大きな手戻りが発生します。
PatSnapはこれを「攻め」と「守り」の二段構えで解決します。攻めを担うEureka R&Dは、特許・論文を横断的に分析して技術課題に対する多様な解決策を提案します。一方、守りを担うEureka IPは、新規性調査やFTO調査を自律的に実施し、知財リスクを早期に発見します。
特に印象的だったのは、AIが単に回答を生成するのではなく、人間のサーチャーが行う検索式作成、反復検索、引用文献評価といった調査プロセスそのものを自律的に実行する点です。これは従来の生成AIによる支援を超えた「AIエージェント型調査」の姿といえます。
特許調査業務は、①人が検索し人が読む従来型、②生成AIが一次処理し人が確認する支援型、③AIが調査プロセスを自律実行し人が監査・最終判断するエージェント型へ進化しつつあります。PatSnapのデモは、その未来像を具体的に示すものであり、「AIに質問する時代」から「AIに調査を任せる時代」への移行を強く感じさせる内容でした。
 
【6月5日開催】Patsnap R&D Insight Webinar
https://www.patsnap.jp/event/0605rdwebinar
 
 
PatSnap’s Vision of AI-Agent-Based Patent Search and Technology Discovery
Recently, I attended a seminar presented by Mr. Sato of PatSnap titled “Toward the Fastest Development Cycle with No Rework.” While the theme focused on rare-earth-free motor development, the essence of the presentation was the transformation of R&D and intellectual property activities through AI agents.
The presentation explained that multiple technological pathways exist for achieving the goal of rare-earth-free motors, and that the choice of pathway can significantly affect development time, cost, and probability of success. In many development projects, teams move into design and prototyping before conducting sufficient prior-art and technology investigations. As a result, patent barriers or technical challenges are often discovered later, causing substantial rework and delays.
PatSnap addresses this challenge through a two-pronged approach of “offense” and “defense.” On the offensive side, Eureka R&D analyzes patents and scientific literature across domains to propose a wide range of solutions to technical problems. On the defensive side, Eureka IP autonomously conducts novelty searches and freedom-to-operate (FTO) analyses, enabling the early identification of intellectual property risks.
What I found particularly impressive was that the AI does not merely generate answers. Instead, it autonomously executes the research process itself—including search query generation, iterative searching, and citation evaluation—the same activities traditionally performed by skilled human searchers. This represents a form of “AI-agent-based investigation” that goes well beyond conventional generative AI assistance.
Patent search and analysis work is evolving through three stages: (1) the traditional model, where humans search and humans read; (2) the AI-assisted model, where generative AI performs preliminary processing and humans review the results; and (3) the agent-based model, where AI autonomously executes the investigation process while humans serve as auditors and final decision-makers.
The PatSnap demonstration provided a concrete glimpse of this future. More importantly, it highlighted the transition from an era in which we “ask AI questions” to an era in which we “delegate investigations to AI.”

Your browser does not support viewing this document. Click here to download the document.

Your browser does not support viewing this document. Click here to download the document.