Overview of Three Field Articles on AI Operations as of the End of June 2026
Operational Patterns Running Through "AI Promotion Specialist," "Claude Code Workflow Design," and "Personal Search Infrastructure Hister"
Target Articles
Zenn / BABY JOB Development Department "4 Months as an AI Promotion Specialist: The Answer Reached Through Trial and Error" (2026-06-26)
Qiita / Haruka Mizuki "I'll Be Honest. You're Using Claude Code Wrong" (2026-06-25)
GIGAZINE "General-Purpose Search Engine Hister" Review (2026-06-27)
Table of Contents
Introduction: The Aim of This Paper and the Three Articles.............................................................. 3
Chapter 1: Zenn "4 Months as an AI Promotion Specialist" — The Layer of People and Organizations.................... 4
Chapter 2: Qiita "You're Using Claude Code Wrong" — The Layer of Work and Design. 6
Chapter 3: GIGAZINE "Hister" — The Layer of Memory and Evidence......................... 8
Chapter 4: The Three-Layer Model — A Single Thread Running Through All Three........................................... 10
Chapter 5: Operational Patterns — Observation, Standardization, Evidence, and Revision.................................. 11
Chapter 6: Translation into SOX/J-SOX Business Flows.............................................. 12
Chapter 7: Connection to Social Observatory, Sekisho, and Vault..................... 13
Chapter 8: The Boundary Between Monitoring and Support — The Subject Problem of Logs...................................... 14
Chapter 9: Corrections and Points to Note........................................................................... 15
Conclusion: A Snapshot as of the End of June 2026............................................ 16
Introduction: The Aim of This Paper and the Three Articles
In late June 2026, within almost the same week, three articles regarding the practical operation of AI agents were published in succession. These are Zenn's "4 Months as an AI Promotion Specialist: The Answer Reached Through Trial and Error" (June 26), Qiita's "I'll Be Honest. You're Using Claude Code Wrong" (June 25), and the GIGAZINE review of the personal search engine "Hister" (June 27). With different titles, media outlets, and target audiences, they appear unrelated at first glance. However, when read side-by-side, it becomes clear that they are answering the same question—what truly works in the era of AI agents—from different entry points.
The first article is an organizational theory piece reflecting on four months of field activity by a specialist in charge of AI promotion. The second article is a work design theory for Claude Code derived from three months of practical experience by a personal developer. The third article introduces a tool that saves viewed pages and local files to make them searchable. In terms of roles, the first deals with people and organizations, the second with work and design, and the third with memory and evidence. These three correspond exactly to the three layers necessary when incorporating AI into business operations.
This paper has two objectives. First, to concisely organize the arguments of the three articles as verifiable facts. Second, to extract the "operational patterns" that emerge in common from them and translate them into more concrete design challenges, such as business flow design compliant with internal controls (SOX/J-SOX), Social Observatory pipelines, Sekisho exit management, and personal knowledge bases centered on Obsidian Vault. In each chapter, I will consciously distinguish between content that can be clearly read from the articles and the interpretations and implications added to them. The goal is not to create a clean summary, but to draw a line through the three and break them down into implementable principles.