I started assigning work at night and making decisions in the morning. Using Dots showed me what it is like to work with a personal assistant and an AI team.
OBSERVATION ARCHIVE
Notes
An English index of notes on AI, UX, marketing, implementation, work, and personal knowledge archives.
English pages are being added one by one without using translation APIs. Only completed English article pages are listed here.
An experiment in letting a top AI handle market research through product creation revealed the gap between a perfect self-audit and a product buyers could not use.
A record of several months of using AI conversations as article material, implementing them with Codex, and growing a personal portfolio after my profile appeared first in a search for my name.
I also have Claude Code at home, but moving the setup and working context built up in Codex takes preparation. A service outage halted article production while I waited for recovery, revealing my dependence on one AI.
Instead of trying to remember everything, keep an index that leads back to original information and use AI as a guide. A practical design for external memory, retrieval, and decision-making.
Choosing what I want to do over invitations. A personal note on distinguishing solitude from being alone and enjoying my own time.
Turn short bursts of intense focus into lasting value. A project design for people who get bored easily: ship small, preserve reusable assets, and make it easy to pause or resume.
An implementation record of a local dashboard that joins Shopify orders and customers, a product master, and Huckleberry subscription orders and contracts to track sales, gross profit, F2, subscription LTV, retention, churn, and data quality.
A two-day redesign that connected articles, tools, implementation case studies, and a profile into an operating system for knowledge integration and practice.
A practical framework for testing AI-generated business ideas against real competitors, substitutes, paying customers, data access, operating costs, and small falsifiable experiments.
A conceptual model for AI outcomes as the product of five factors: model capability, unique context, question quality, review skill, and learning speed. Use it to find the bottleneck in your AI workflow.
Treat generative AI as a capable new hire who does not know your organization. Externalize your decision criteria so you can move from detailed task instructions to delegating outcomes.
An implementation case study of a local dashboard that loads weekly ad data, separates messaging angles, keywords, and product targets, tracks ten metrics, and suggests what to test next.
Why AI efficiency does not always reduce work. Based on e-commerce operations, this article explains how saved time can expand an individual's scope, hidden responsibilities, and maintenance burden.
A reflection on building a price-comparison web service in about four working days. The main time savings came not just from code generation, but from reducing handoffs and coordination between roles.
How to use AI to increase the number of learning experiments, turn real work into practice, and keep expanding what you can do after 40.
A UX case study of building a fictional government agency that selects visitors as candidates and puts them through five useless tests, from concept and copy to UI, development, and launch in one day.
A case study of planning, designing, illustrating, building, and publishing Wish, Made in Japan., a gift site for overseas shoppers, in two days by myself.
Why running multiple AI agents shifts the bottleneck from execution to human review—and how to manage review queues, source-of-truth files, task states, and safe concurrency.
A case study of Career Turning Point Journal, a free assessment built in about two days. It brings career questions and fortune-telling into one UX, from concept and diagnostic logic to launch and measurement.
A case study of building an MTG sealed-product price comparison site in about five days with ChatGPT and Codex. It covers matching 150 products, daily API updates, price history, UX, SEO, and security.
An implementation note on building an AI control room that automatically consolidates progress, execution evidence, pending approvals, project records, and lessons from failures across multiple AI and Codex projects.
How to recognize a workplace where capable employees absorb extra work without recognition, and decide when to improve the situation or move on.
Experience in marketing, e-commerce, legal review, data operations, and AI can support a move into AI process improvement or digital transformation. This note shows how to frame that transition.
An AI note on turning small daily inconveniences into tools and reusable work assets, and the strengths of builder-oriented talent.
Why organizations that reward long hours and tolerance for manual work overlook process improvers, and what to reward in the AI era.
How non-engineers can turn frontline CSV work, matching, checks, masking, and reporting into small AI-assisted tools—and make process improvement systematic.
A practical design for safer AI agents: let them research, compare, draft, and organize risks, then stop for human approval before sending, purchasing, posting, or signing.
A method for breaking complaints about a UI or tool down by individual actions to identify improvements and opportunities for small tools.
A practical safety design for running Codex and AI agents with broad permissions: use a clean, isolated PC, disable cloud sync, rotate exposed credentials, and prepare recovery options.
Why practical AI results depend on more than prompt-writing: build a reusable reference base of lessons, success and failure factors, constraints, and decision criteria from real work.
A practical note on treating workplace criticism and reprimands as data about differences between your own judgment standards and someone else’s, rather than as emotional events alone.
An implementation note on safely testing AI tools and business improvement ideas by abstracting work processes instead of using company data or personal information.
A practical note on redefining an employee’s role as reducing work and building systems that prevent it from recurring, rather than simply taking tasks on.
A practical guide to internal-tool quality: prevent incorrect results, make workflows easy to understand, and build for reliable operation instead of adding features for their own sake.
A practical note on treating KPI creation, signal detection, and progress reviews as decision design that helps project managers move work forward, not merely as reporting.
A practical way to distinguish work that requires a specialist from tasks a dedicated tool can handle, making in-house work more feasible in the AI era.
A practical argument for treating repetitive copying, conversion, exports, and routine checks as tasks to eliminate rather than work that needs human attention.
A way to use AI to turn small daily or workplace inconveniences into tools and treat them as seeds for products that can be published.
An AI note on freezing specifications, defining prohibited changes, and reviewing diffs before asking Codex to faithfully port an existing tool or HTML.
An AI note on explaining the impact of reducing 30–60-minute manual tasks to seconds using time saved, task frequency, and the number of people affected.
Why developing with AI agents or Codex builds practical skills across requirements, UX, acceptance, operations, and security.
AI development improves when implementation, review, design, and auditing are assigned to distinct roles instead of one general-purpose agent.
Practical guidance on the benefits of building internal HTML and JavaScript tools without CDNs, external network requests, or non-local dependencies.
A self-analysis from a project management perspective: the practical strength is not just numerical skill, but the ability to read gaps and inconsistencies as structure.
How a Chrome extension can extract visible page content or posts and pass them to AI for analysis, even when an API is unavailable.
Reframing hover, scrolling, cards, and links as UX responses that return meaning and atmosphere when people interact, rather than as mere decoration.
E-commerce optimization goes beyond product pages and ads. Pricing, delivery, email, and CRO need to work together for both customer experience and profitability.
Rethinking a personal site as a public working asset that demonstrates AI use, UX, writing, and implementation—not merely a place to post a profile.
An AI note on designing chatbots around concrete user problem solving instead of treating conversation itself as the value.
A note on evaluating AI subscriptions as production infrastructure, saved time, and reusable intellectual assets rather than mere tool spending.
A work note on starting AI adoption from decomposing work into tasks, decisions, checks, and exceptions rather than from prompt training.
A marketing note that reframes marketing not as persona creation or campaign execution, but as designing the conditions that raise purchase response probability.
- Build a Judgment OS, Not a Memo Pile: Why an Insight Database Becomes a Personal Asset in the AI Era
An AI note that reframes an insight database as a personal OS that turns judgment criteria, prohibitions, article ideas, and AI instructions into reusable context.
An AI note on treating Codex requests as constraint design, stabilizing implementation quality by giving purpose, scope, prohibitions, and completion criteria.
A marketing note that reframes Deep Research as a personal research department for hypotheses, competitor analysis, article development, and strategic decisions.
An AI note on why strength now comes from connecting problem definition, design, AI direction, verification, and operation rather than raw task speed.
An AI note on starting agent adoption from evaluation criteria, context design, permissions, and role boundaries before choosing tools.
An AI note on failure patterns, evaluation, verification, and human-side review design.
An implementation note on remote desktops, AI agents, instruction design, and verification design.
A UX note on reducing hesitation, anxiety, and effort so users can naturally move to the next action.
Treat AI not as a one-off time-saver but as a compounding system: turn conversations into lessons, implementations into criteria, and articles into reusable knowledge that improves future work.
A practical approach to AI-agent safety based on recoverability, isolated environments, limited permissions, and human review—not blind trust.
A UX note on AI intervention timing, transparency moments, and the limits of chat UI.
A note on designing AI agents around roles, data access, decision authority, and evaluation criteria rather than personality.
A UX note on participant design, exclusion criteria, and user panel decay before speeding research up with AI.
A note on turning insight logs into judgment criteria, output rules, and prohibitions that a future personal AI can use.
A note on designing information collection as an automated knowledge pipeline.
A marketing note on designing content that can be selected, cited, and trusted by AI search experiences such as ChatGPT, Perplexity, and Gemini.
Designing Notion as a structured knowledge database that AI can reference and reuse.
Why reusable judgment criteria matter more than simply accumulating information.
How AI shifts web production value toward research, UX design, operations, and outcome design.
Why AI becomes more useful when it reads your own products, customers, failures, constraints, and judgment criteria.
Turning advanced global insights in AI, UX, marketing, and EC into practical Japanese business operations.
How broad practical experience becomes valuable when coordinating AI, tools, information, and specialists into outcomes.
An implementation log on combining GitHub Actions, Claude API, and Notion API to collect AI and marketing information automatically.
Growing a personal site as a hub that connects observations, articles, achievements, and external profiles.
Why personal data such as judgment criteria, observations, failures, and success patterns should be accumulated before agents become mainstream.
Building reusable AI context through automated information retrieval, structuring, and storage.
PDP design for AI search, AI shopping, and agent-based discovery, including structured data, reviews, and FAQ design.
Email as a controllable customer touchpoint when AI search, social algorithms, and ad costs make external channels unstable.
A UX case study on why people who use AI deeply can feel isolated and need communities that understand their working temperature.
The stress caused when the abstraction level required by work does not match the way a person naturally thinks.
Why past logs should be treated as summaries of past judgment criteria, not as a replacement for present judgment.
How company rules, culture, and constraints around AI use can affect experience accumulation and future market value.
Using ChatGPT not as an article generator but as a thinking partner whose conversation logs become article ideas and portfolio material.
Using AI to standardize compliance checkpoints, prohibited expressions, correction candidates, and decision grounds across companies.
Why RAG is needed when internal prohibited expressions, past review comments, and approval lines cannot be judged by general AI alone.
Extracting observations from ChatGPT conversations, placing them into templates, and publishing through GitHub and Cloudflare Pages.
How human value concentrates in defining the starting conditions and judging the output as AI accelerates the middle process.
Why faster reasoning modes reduce the reason to choose instant answers when using ChatGPT as a thinking support tool.
How making, publishing, and accumulating work logs become proof of individual value.
The limits of the idea that adding context always improves accuracy, and why field problems are harder than they look.
How the human role shifts toward judgment, design, and responsibility when AI can create drafts.
The domains where human value remains by staying close to change, context, and specialized practical judgment.
Shifting content operations from making polished pieces to continuously streaming thinking as a byproduct.
How AI reduces tasks rather than work itself, increasing the importance of judgment and process design.
The shift from humans thinking and AI assisting to AI handling analysis and decision support while humans design and execute.
Throwing rough ideas to AI, detecting misalignment, and converging through loops instead of expecting the first answer to be correct.
Solving the problems of invisibility, comparison difficulty, and manageability through a management UI.
Why value comes from making excellent open-source software usable rather than building everything from zero.
The difference between compressed AI and structured AI, and why fit with personal thinking style matters.
How AI can process multidimensional data without forcing it into two-dimensional human-friendly views.
How work speed changes when people design questions, judge quickly, and turn work into systems.
How the human role in marketing shifts to designing structures that AI can analyze well.
Turning knowledge into reusable assets through accumulation, structuring, and external publication.
Using AI as a converter that turns thinking notes into publishable content with almost no psychological burden.
A portfolio note based on the Google UX design framework, covering empathy research, personas, journey maps, problem definition, user flow, validation, and information architecture.
Reconnecting persona, customer journey, advertising, landing pages, and UX through the principle that people perceive after predicting.
A portfolio note organizing the UX design process for a pet dental care product.
How professional knowledge and judgment criteria change when they can be templated and externalized.
An AI-era development mindset that starts with selecting existing OSS and filling only the missing parts with AI.
Abstracting the decision logic of capable people and moving it into AI as a structure that does not depend on one individual.
The strength of combining multiple skill domains rather than relying on a single replaceable task.
The thinking process and practical knowledge for designing a site without relying on vague taste.