Daily AI Zine
Wednesday, September 16, 2026
Issue No. 061 · Tokyo · full edition

✦ Today's Big Thing

Cloudflare redraws the search versus AI-training line

A quiet but useful shift: site owners can stay visible in search while refusing AI training, which turns content policy into an operational setting rather than a vague legal footnote.

5 min read · 7 sections

In Brief
  1. Cloudflare now gives site owners controls to stay discoverable in search while disallowing AI training.
  2. GitHub Copilot can suggest allowed values when creating repository custom properties.
  3. OpenAI’s Agents API turns long-running Codex-style work into a managed cloud service.
  4. Hot signal: Meta One packages higher AI usage and creator tools across Meta’s apps.

Today's Big Thing

The one thing that matters

Use it

Cloudflare separates search visibility from AI training

The web is getting a more practical consent layer.

Cloudflare is giving site owners a way to stay discoverable in search while disallowing AI training. The important part is the split: search indexing and model training no longer have to be treated as one bundle. Cloudflare also says the controls use an Accountable designation with Apple, Google, and Microsoft, which makes this more than a private dashboard toggle. For any operator with public sites, guides, case studies, or production know-how online, this is a useful shift. It does not settle the broader licensing fight, but it creates an immediate website-operations task: decide which content should be visible to customers and which content should not be training material.

T1Cloudflare Blog09/15

My AI Ecosystem

Your actual stack

Test it

GitHub Copilot starts helping with repository metadata

GitHub Copilot can now suggest allowed values when creating a custom property for repositories in an organization. It is in public preview for GitHub Copilot Business and Copilot Enterprise. This is admin plumbing, but useful: cleaner repo labels make agent routing, ownership, review rules, and portfolio dashboards less manual.

T1GitHub Changelog09/15
Test it

OpenAI’s Agents API makes Codex-style cloud work operational

OpenAI introduced the Agents API as a managed service for cloud agents, powered by the Codex harness for orchestration, long-running sessions, and tool use. The practical change is delegation: tasks that are too long for a normal chat can become monitored cloud work with tools attached. Test it only on reviewable, low-risk workflows first.

T1OpenAI News09/10
Watch it

Hot signal: Meta One packages AI creation for businesses

Meta introduced Meta One, a subscription service across Facebook, Instagram, WhatsApp, and Meta AI with more AI usage, expression features, and tools for creators and businesses. This is not a single breakthrough feature. It is a distribution signal: AI creation tools are being sold where small brands already post, message, and sell.

T1Meta Newsroom09/15

CoWork Corner

Claude CoWork, day to day

CoWork: no product move, but containment still matters

Claude CoWork is the workspace Adrian uses for persistent projects, operational records, and AI-assisted workflows. No meaningful CoWork product change surfaced in today’s candidates. The useful reminder comes from Anthropic’s current engineering focus on containing Claude across products as agents become more capable and their possible blast radius grows. Revisit one workflow: keep CoWork as the place for project memory, decisions, and handoff records, while executable work remains in a separate coding or MCP lane with explicit review before files, deploys, or external messages change.

Tier 4 · quiet day, honest fallback

T1Anthropic Engineering Blog

GPT Desk

OpenAI, ChatGPT, Codex

Test it

Put the Agents API on one reviewed operations job

OpenAI changed the build surface with the Agents API: cloud agents can now run long sessions, use tools, and operate through a managed Codex harness. Adrian should care because Grey Group OS needs repeatable jobs that do not depend on a fragile chat thread. Today, define one low-risk task for it: take a prepared project brief and produce a reviewed task list, owner assumptions, and missing-input checklist before any client-facing output exists.

T1OpenAI News09/10
Use it

Turn Codex work into an experiment log, not a coding blur

OpenAI says coding agents are reshaping internal research by affecting agent usage, experiment velocity, task complexity, and research acceleration. Adrian should care because Goodsense, SET, and internal product work all need faster learning without losing why a decision was made. Today, make every Codex run produce one short experiment record: goal, files touched, result, failure, next test. That turns AI coding into reusable operating memory.

T1OpenAI News09/06

Small Money Systems

Small, repeatable, real

Use it

Small system: AI-training policy mini-audit

The System is a one-page website AI-access audit that checks whether a public site can stay searchable while refusing AI training. The Customer is a Japan-facing founder, small agency, school, consultant, or production vendor with valuable public content. The Offer is a short policy note, recommended Cloudflare setting path, and a before-and-after content-risk summary. The Price is ¥35,000. The Existing assets are Grey Group OS checklists, Netlify and GitHub site experience, and Adrian’s Japan-English operator judgment. The AI workflow uses Claude Code or Codex to inspect site structure and draft the audit note for human review. The First action is to make a two-column checklist for search visibility and AI-training permission. The Repeatability comes from reusing the same audit template across sites. The Effort is one hour. The Expected value is small service revenue plus leads for broader site and workflow cleanup.

T1Cloudflare Blog09/15
Test it

Small system: Meta One creator test pack

The System is a small Meta AI creator-workflow test pack for local brands and event teams. The Customer is a venue, retailer, creator, or activation partner already posting on Instagram, Facebook, or WhatsApp. The Offer is one tested content workflow, five reusable post directions, and a simple decision on whether Meta One is worth paying for. The Price is ¥45,000. The Existing assets are Street Attack Japan activation thinking, Goodsense creative systems, and Adrian’s production eye. The AI workflow uses ChatGPT for planning, Claude Code for packaging the template, and manual review for tone and rights. The First action is to create a one-page intake for brand, audience, offer, and posting cadence. The Repeatability comes from turning each test into a reusable vertical template. The Effort is half day. The Expected value is paid discovery work that can lead to monthly content support.

T1Meta Newsroom09/15

Build Next

Deployable now

Use it

Build a lightweight AI-access policy checker

What to build is a small GitHub-backed checklist page for sites that records search visibility, AI-training permission, crawler notes, and last review date. Why now is Cloudflare’s new control path for staying discoverable while disallowing AI training. Effort is one hour. Expected impact is clearer content governance and less repeated policy discussion before publishing public material. Dependencies are an existing site, access to its hosting or Cloudflare settings, and a simple repository or notes page to store the decision trail.

T1Cloudflare Blog09/15

Try This Today

One action, right now

Pick one public site or content library. Add a single row to your ops notes: should this stay searchable, should it allow AI training, who owns the decision, and when it gets reviewed. That row is enough to start the policy habit.