Daily AI Zine
Saturday, September 12, 2026
Issue No. 058 · Tokyo · full edition

✦ Today's Big Thing

Agent work is moving into the measurement layer

GitHub’s newest Copilot metrics update is the practical signal today: agents are no longer just something teams run, they are something teams can track.

5 min read · 7 sections

In Brief
  1. GitHub Copilot usage reports now include activity from the VS Code Agents window.
  2. Copilot’s weekly release adds Jira integration, adaptive model routing in the CLI, and more VS Code agent automation.
  3. OpenAI’s Codex and ChatGPT case study points to AI as a discovery workflow, not only a coding workflow.
  4. Cloudflare’s automatic remediation policies show security work shifting from alerts into action rules.

Today's Big Thing

The one thing that matters

Use it

Copilot agents become measurable work

The useful shift today is not another model. It is a way to see whether agent work is actually being used.

GitHub Copilot usage metrics reports now include generally available metrics for activity in the dedicated VS Code Agents window. That matters because agent adoption is moving from personal experimentation into team operations. If a team cannot see who is using agents and where the activity sits, it cannot manage training, cost, permissions, or workflow quality. Treat this as a control-layer update: first measure agent use, then decide which work belongs with agents. The immediate move is to start a weekly agent activity snapshot for active repositories and compare it with shipped tasks, not chat volume.

T1GitHub Changelog09/11

My AI Ecosystem

Your actual stack

Test it

Copilot’s weekly release adds workflow plumbing

GitHub’s September 7 Copilot release adds Jira integration in the Copilot app, adaptive model orchestration with Project HydraFusion in Copilot CLI, and new agent automation in Visual Studio Code. The practical point is routing: tickets, model choice, and IDE agent work are being pulled closer together. Test it on one non-critical repo before changing the default coding workflow.

T1GitHub Changelog09/10
Watch it

OpenAI shows the scale behind ChatGPT reliability

OpenAI says it evolved Habitat from a Python library into a globally distributed storage platform serving over 1 billion ChatGPT users and 22 million requests per second. This is not a feature to adopt, but it is a reminder that persistent AI products depend on storage design as much as model choice. Watch for reliability and data-retention implications in API-backed workflows.

T1OpenAI News09/11
Watch it

Hot signal: Cloudflare automates SaaS risk remediation

Cloudflare introduced CASB automatic remediation policies, a native automation engine for SaaS risks. Security teams can design event-driven logic to revoke risky file shares and send webhooks without manual intervention. This is a useful signal for agent operations: the market is moving from dashboards toward approved action rules. Test the pattern conceptually before giving any AI system write access.

T1Cloudflare Blog09/11

CoWork Corner

Claude CoWork, day to day

CoWork: make executable work a separate MCP lane

Claude CoWork is the workspace Adrian uses for persistent projects, operational records, and AI-assisted workflows. Anthropic’s engineering index continues to surface code execution with MCP as an agent-efficiency topic, which is enough to justify one workflow rule: keep CoWork as the planning and memory surface, then move executable actions into a named MCP lane with a narrow task, clear input files, and a written stop condition. Do not let a general project conversation become the place where commands, edits, and decisions blend together.

T1Anthropic Engineering Blog

GPT Desk

OpenAI, ChatGPT, Codex

Test it

Use Codex as a discovery partner, not only a builder

OpenAI says César de la Fuente’s lab uses Codex and ChatGPT to search living and extinct genomes for antimicrobial candidates. The change is the pattern: Codex is being used to navigate a large technical search space, while ChatGPT helps shape the research workflow. For Adrian, that maps to Grey Group OS, Goodsense, SET, and Japan-English proposal research. Today, give Codex one folder of prior decks or notes and ask for a source-cited opportunity map, then have ChatGPT turn it into a one-page brief.

T1OpenAI News09/10
Use it

Make OpenAI work carry an owner line

OpenAI’s Gilbert + Tobin case study centers on CEO-led commitment, rigorous governance, and human accountability while scaling ChatGPT Enterprise and Codex. The useful lesson for Adrian is simple: AI output should not enter Grey Group OS, Goodsense, SET, or Street Attack Japan without an accountable human owner. Today, add one required field to every AI-produced brief or code task: owner, review status, and where the final decision lives.

T1OpenAI News09/01

Small Money Systems

Small, repeatable, real

Use it

Small system: agent usage snapshot pack

System: build a one-page monthly agent usage snapshot for small teams using GitHub Copilot. Customer: Japan-based agencies, startups, or production-adjacent tech teams paying for Copilot but not measuring agent work. Offer: they receive a short report on VS Code Agent activity, adoption gaps, and next workflow rules. Price: ¥55,000 per snapshot. Existing assets: Grey Group OS, GitHub practice, Claude Code, Codex, and proposal workflow experience. AI workflow: Codex drafts the report shell, Claude Code builds the reusable template, and ChatGPT polishes the client version. First action: create the report outline from GitHub’s new Copilot metrics fields. Repeatability: the same template runs monthly. Effort: one hour. Expected value: small recurring revenue and a clean lead-in to AI workflow audits.

T1GitHub Changelog09/11
Test it

Small system: Jira-to-agent ticket cleanup

System: sell a lightweight cleanup service that turns messy Jira tickets into agent-ready implementation cards. Customer: teams that already use Jira, GitHub, and AI coding tools but lose time on vague requests. Offer: ten cleaned tickets with acceptance criteria, risk notes, and the right agent lane. Price: ¥88,000 for the first batch. Existing assets: Grey Group OS delivery rules, Codex, Claude Code, GitHub, and production planning habits. AI workflow: ChatGPT rewrites the business intent, Codex checks implementation shape, and Claude Code can create the reusable card generator. First action: draft one before-and-after ticket sample today. Repeatability: each client can buy another batch before a sprint. Effort: half day. Expected value: paid cleanup work that can become a monthly delivery-support retainer.

T1GitHub Changelog09/10

Build Next

Deployable now

Use it

Build a Copilot agent activity page

What to build: a simple internal page that records weekly VS Code Agent activity from Copilot usage metrics beside shipped repository work. Why now: GitHub now includes generally available metrics for the dedicated VS Code Agents window. Effort: one hour. Expected impact: better decisions on where agent work is useful, where it is wasted, and which repos need rules. Dependencies: access to Copilot usage metrics, a GitHub repo list, and a place to publish or store the weekly snapshot.

T1GitHub Changelog09/11
Use it

Build a Jira-to-agent task card template

What to build: a reusable task card that turns a Jira issue into agent instructions, required files, acceptance criteria, review owner, and stop condition. Why now: Copilot’s weekly release adds Jira integration in the Copilot app, making ticket-to-agent workflow more immediate. Effort: one hour. Expected impact: less rework when handing implementation tasks to Codex, Claude Code, or Copilot agents. Dependencies: a live Jira-style issue, an active repo, and one reviewer who can approve the final task card.

T1GitHub Changelog09/10

Try This Today

One action, right now

Open the repo you touched most this week and add one row: task, tool used, agent window activity if available, human reviewer, shipped result. Do this once before adding more automation. The metric only matters if it ties agent use to delivered work.