Daily AI Zine · No. 047

Today's Big Thing

The one thing that matters

Use it

Polimill turns municipal knowledge into AI infrastructure

Japan’s most useful AI signal today is not a new model. It is a public-sector operating pattern.

OpenAI says Polimill is using GPT models and Codex to help municipalities search and use administrative knowledge while accelerating development. That matters because local government knowledge is usually fragmented, procedural, and hard to reuse across departments. The practical shift is clear: AI value comes from making institutional memory searchable, then using a coding agent to keep the system improving. This is a strong Japan signal because it shows GPT and Codex moving into public administration, not just software teams. Verdict: use the pattern, not the exact sector. Any organization with recurring approvals, bilingual documents, or procedural handoffs can copy the structure at small scale.

T1OpenAI News08/31

My AI Ecosystem

Your actual stack

Test it

GitHub Copilot gets better session control in VS Code

GitHub’s August VS Code Copilot releases focus on organizing agent sessions, reviewing changes, and navigating long conversations. The useful part is not another chat feature. It is the move toward keeping agent work inspectable after the first prompt. If a repo task now spans multiple turns, treat the session as a work artifact that needs naming, review, and cleanup.

T1GitHub Changelog08/31

Test it

Hot signal: Runway makes video model quality the next test

Runway lists Gen-4.5 as a video model focused on motion quality, prompt following, and visual fidelity. That is worth attention for commercial production, but only through a controlled test. Run one internal comparison: same product shot, same brief, same timing, measured against editability and client usefulness. If it cannot survive revision notes, it stays a reference tool.

T1Runway Research

Use it

Codex keeps moving toward business builders

OpenAI’s loveholidays case says Codex is being used to make software development accessible across the business, not only inside engineering. The practical takeaway is a routing rule: use Codex for small internal tools where the domain owner can clearly describe the workflow and verify the result. Do not reserve it only for formal product engineering.

T1OpenAI News08/26

Test it

MCP code execution remains the agent efficiency pattern

Anthropic’s engineering index points to code execution with MCP as a way to build more efficient agents. In plain terms, MCP lets an AI tool reach approved external capabilities, and code execution lets it test work instead of only describing it. The immediate use is narrow: attach execution only to tasks where a pass or fail result is visible.

T1Anthropic Engineering Blog

CoWork Corner

Claude CoWork, day to day

CoWork stays steady, revisit the handoff note

Claude CoWork is the workspace Adrian uses for persistent projects, operational records, and AI-assisted workflows. No meaningful CoWork product change appears in today’s candidates. The useful revisit is simple: before moving a task from discussion to execution, write a handoff note with goal, constraints, source files, success check, and stop condition. Anthropic’s Claude Code best-practice material keeps pointing in the same direction: agentic work improves when the task boundary and verification path are explicit.

Tier 4 · quiet day, honest fallback

T1Anthropic Engineering Blog

GPT Desk

OpenAI, ChatGPT, Codex

Use it

Turn Polimill into a Japan knowledge-search demo

What changed: OpenAI says Polimill uses GPT models and Codex to help municipalities search and use administrative knowledge while speeding development. Adrian should care because Goodsense, SET, and Grey Group OS all benefit when Japan-English institutional knowledge becomes a reusable briefing layer instead of scattered notes. Do this today: make a 20-record demo index from public municipal or market-facing materials, then ask Codex to scaffold a tiny search-and-summary interface around it.

T1OpenAI News08/31

Test it

Give Codex an ops owner, not a ticket queue

What changed: OpenAI’s loveholidays story frames Codex as a way for people across a business to turn ideas into products faster. Adrian should care because Grey Group OS needs small internal utilities more often than it needs large software projects. Do this today: choose one non-engineering workflow, such as proposal version tracking or post-shoot asset intake, and give Codex one written owner brief, one acceptance test, and one GitHub issue.

T1OpenAI News08/26

Small Money Systems

Small, repeatable, real

Test it

Small system: municipal briefing snapshot

System: build a paid Japan municipal or local-market briefing snapshot that turns public administrative knowledge into a short bilingual opportunity brief. Customer: foreign producers, agencies, or founders preparing a Japan project. Offer: a 5-page snapshot with relevant offices, process notes, terminology, risks, and next-step contacts where publicly identifiable. Price: ¥30,000 to ¥50,000. Existing assets: Goodsense research habits, SET production planning, Japan-English workflow, and Grey Group OS templates. AI workflow: ChatGPT drafts the structure, Codex builds a reusable intake and briefing template, and Claude Code can refine the repository. First action: pick one ward or city and create the first sample brief today. Repeatability: each location becomes a reusable record. Effort: half day. Expected value: lead generation plus small paid research work.

T1OpenAI News08/31

Use it

Small system: ops intelligence retainer

System: sell a lightweight monthly ops-intelligence room that turns scattered client notes, campaign updates, and production decisions into one searchable brief pack. Customer: small agencies, production partners, or Japan-facing teams that lose time finding past decisions. Offer: monthly intake cleanup, searchable summaries, decision log, and next-action sheet. Price: ¥80,000 per month. Existing assets: Grey Group OS, Daily Ops discipline, Goodsense proposal work, and production coordination templates. AI workflow: ChatGPT summarizes, Codex builds the intake and export flow, and GitHub stores versioned templates. First action: create a one-page sample using an internal dummy project. Repeatability: the same room structure renews monthly. Effort: one hour for the sample, larger build for automation. Expected value: recurring revenue or time savings for a retained client.

T1OpenAI News08/12

Build Next

Deployable now

Use it

Build a tiny admin-knowledge search demo

What to build: a small searchable knowledge demo with 20 public records, a summary view, and a source-note field. Why now: OpenAI’s Polimill case shows GPT models and Codex being used for municipal administrative knowledge, making the pattern timely for Japan-facing work. Effort: half day. Expected impact: faster briefing output and a concrete sales demo. Dependencies: a Codex account, a GitHub repo, approved public source material, and one narrow use case.

T1OpenAI News08/31

Use it

Build an agent-session review card

What to build: a markdown review card for AI coding sessions with task name, files touched, tests run, unresolved questions, and human approval. Why now: GitHub’s VS Code Copilot releases emphasize organizing agent sessions, reviewing changes, and navigating longer conversations. Effort: 15 minutes. Expected impact: cleaner code review and fewer lost decisions. Dependencies: an active GitHub repo and a rule that every agent session leaves a short review artifact.

T1GitHub Changelog08/31

Try This Today

One action, right now

Take one public Japan-facing process you already understand and turn it into a single structured record: title, source note, summary, decision points, glossary, next action. If the record is useful, make 19 more and let Codex scaffold the demo.