[From Just an Idea to a 'Working Prototype' in a Day] The Tangible System Development Experience Brought by "AI Agents"

Giken Shoji International Co., Ltd. (Nagoya HQ: Nagoya, Aichi; Tokyo HQ: Shinjuku, Tokyo; CEO: Tomomi Kojima; hereinafter "the company"), a leading domestic company in area marketing GIS (Geographic Information System), has introduced a new development method in the development process of its location intelligence platform "THE NOVEL," which involves running three types of generative AI in parallel. Under the leadership of Project Manager (PM) Wataru Koshiguchi, this method has made it possible to create a web mock-up (including production data and calculation logic), which usually takes several days, in just a few hours, thereby reducing the "rework risk" caused by communication gaps among stakeholders in the upstream development process. 〇 A Key Change: Reversing the Process of Requirements Definition and Development The most significant feature of this initiative is the reversal of the traditional "requirements definition (agreement) -> development" sequence. By having AI generate a mock-up that includes production data and calculation logic in a few hours, a luxurious development process of "solidifying requirements after everyone has touched it" has become possible. 〇 For more details: https://www.giken.co.jp/information/20260518/ ■ Achieving "Upstream Real-Machine Confirmation," Which Was Previously Impossible Due to High Costs - Traditional Structural Issues and AI-Powered Solutions In conventional development, creating a "working prototype (mock-up) with logic" before specifications were finalized was difficult due to cost constraints. This forced reliance on document-based agreements, creating a constant risk of major rework when the final product turned out to be "different from the initial image." To solve this problem, we established a method to generate a prototype screen with calculation logic at an extremely low cost and in a short time by leveraging AI agents. This enabled discussions based on "actual user feel" from the specification review stage, making it less likely for communication gaps to occur. - Background: Overcoming a "One-Month Stall" in the Development of 'THE NOVEL' In the planning phase of the new product, the demands of the sales, engineering, and marketing departments conflicted, causing discussions to stall for about a month. Even with logical frameworks, an agreement could not be reached due to discrepancies in the "interpretation of preconditions." To break through this "verbalization barrier," the project was moved forward by switching to an approach of presenting a "screen with real logic running" without waiting for final specification approval. ■ Concrete Results from Utilizing AI Agents ① "Ultra-Fast, Low-Load Prototyping" with Three Types of Generative AI - Roles such as "design composition" and "use case design" were assigned to ChatGPT, Codex, and Claude Code, which were run in parallel. - The time required for web mock-up creation, which usually takes several days, was reduced to a few hours, establishing a system where feedback from meetings could be immediately reflected as a "working mock-up." ② 50% Reduction and Automation of the Acceptance Testing Process - By linking Claude Code with Playwright, the acceptance testing process, which typically takes two months, was shortened to practically within one month (including the revision period). ③ Transformation of the PM's Role and Optimization of Resource Allocation - The PM's workload was freed from operational tasks such as "communicating specifications." - The freed-up resources were reallocated to a system focused on essential judgments like "customer value consideration" and "edge case decisions," thereby maintaining the purity of the product. ■ Comment from "THE NOVEL" PM Wataru Koshiguchi "Prototype-driven development is a long-standing method. However, by using AI this time, we were able to use data that was just like production data and incorporate business logic using that data into the mock-up. This is more like a final product than a so-called mock-up. I believe we are achieving true agile development. As a result, we can spend more time in the upstream review stage discussing realistic operations, how to explain it to customers, and whether the customer value is essential. The participating members are also making more comments as users rather than just scrutinizing the correctness of the requirements. I feel that the team's creativity has also dramatically increased." - Koshiguchi's Bio Representative of InkField LLC. At Recruit Co., Ltd., he led the launch and popularization of the POS register app "Air REGI" as a PM. He then worked as a development PM for products like the AI translator "POCKETALK" before his current position. He currently serves as the project manager for our system "THE NOVEL." ■ About the Location Intelligence Platform "THE NOVEL" Solution Name: THE NOVEL Category: Location Intelligence Platform Launch: April 2026 *First feature "Sales Forecast AI" Primary Target Departments: Store Development, Corporate Planning, Sales, Marketing URL: https://www.giken.co.jp/products/ai-novel/ ■ About Our Company Provision of GIS for market analysis and area marketing - Market analysis/area marketing GIS "MarketAnalyzer® 5" - Cloud GIS equipped with GPS location data "KDDI Location Analyzer" and others Provision of accumulated analysis data and know-how - Support for advancing data strategy through the provision of data, APIs, and GIS engines - Construction of in-store visitor detection systems - Advancement of digital advertising segmentation Company Name: Giken Shoji International Co., Ltd. (https://www.giken.co.jp/) Representative: Tomomi Kojima, CEO Location: Nagoya Head Office: GSI Building, 2-30 Shuzucho, Higashi-ku, Nagoya, Aichi; Tokyo Head Office: PMO Shinjuku Gyoenmae 7F, 2-1-12 Shinjuku, Shinjuku-ku, Tokyo Established: January 1976 Capital: 231,125,000 JPY