Daily AI Zine
Wednesday, August 26, 2026
Issue No. 041 · Tokyo · full edition

✦ Today's Big Thing

ChatGPT Work and Codex get an admin control surface

Today’s practical lead is workspace control: usage, permissions, limits, and admin requests move closer to the tools doing the work.

5 min read · 7 sections

In Brief
  1. OpenAI added an Admin plugin for ChatGPT Work and Codex to analyze usage, manage members and permissions, adjust limits, and act on admin requests.
  2. GitHub Copilot’s Customize tab is generally available, with MCP support for connecting team tools and workflows.
  3. Hot signal: OpenAI reported first results for its Jalapeño inference chip, pointing to faster, lower-latency model serving.
  4. CoWork has no meaningful product change in today’s candidates, so the useful move is tightening MCP execution rules.

Today's Big Thing

The one thing that matters

Use it

ChatGPT Work and Codex get an admin control surface

The important shift is not a new model. It is governance moving into the workspace.

OpenAI introduced an Admin plugin for ChatGPT Work and Codex that can analyze workspace usage, manage members and permissions, adjust limits, and act on admin requests. That is the useful development today because agent work is no longer just prompting and coding. It is account control, cost control, access control, and handoff control. For any team running ChatGPT Work and Codex side by side, admin work can now sit closer to the work itself instead of being handled as a separate spreadsheet or Slack chase. Verdict: use it as an operating layer, not as another dashboard to admire.

T1OpenAI News08/25

My AI Ecosystem

Your actual stack

Test it

GitHub Copilot customization becomes a real workspace surface

GitHub says the Copilot app Customize tab is now generally available, bringing MCP into the place where teams connect tools, knowledge, and workflows. MCP means Model Context Protocol, a standard way to let AI tools reach approved systems. This matters because customization is moving from prompt craft into team infrastructure.

T1GitHub Changelog08/25
Watch it

Hot signal: OpenAI’s inference chip moves from concept to results

OpenAI reported first results for Jalapeño, its custom inference chip. Inference is the serving step that turns a prompt into an answer. OpenAI says the chip delivers faster, more power-efficient inference with higher throughput and lower latency for modern models. This is not something to build with today, but it is a cost and speed signal for API-heavy workflows.

T1OpenAI News08/25

CoWork Corner

Claude CoWork, day to day

CoWork has no product move, so tighten execution rules

Claude CoWork is the workspace Adrian uses for persistent projects, operational records, and AI-assisted workflows. No meaningful CoWork product change showed up in today’s candidates. The useful revisit is MCP code execution: write one project rule that says which scripts an agent may run, what files it may read, where outputs must be saved, and when it must stop for human review. MCP code execution is useful only when the room has clear boundaries.

Tier 4 · quiet day, honest fallback

T1Anthropic Engineering Blog

GPT Desk

OpenAI, ChatGPT, Codex

Use it

Give ChatGPT Work and Codex an admin lane

OpenAI changed the operating surface by adding an Admin plugin for ChatGPT Work and Codex. Adrian should care because Grey Group OS needs permission, member, limit, and usage checks to be part of delivery, not cleanup after the fact. Today’s action: create one admin review template for Goodsense and SET workspaces with three fields only, active members, Codex usage, and limit exceptions.

T1OpenAI News08/25
Watch it

Put Jalapeño into the API cost-routing watchlist

OpenAI says Jalapeño’s first results show faster, more power-efficient inference, with higher throughput and lower latency. Adrian should care because Grey Group OS, Daily Ops, and proposal automation all depend on picking the right model lane for speed, cost, and reliability. Today’s action: add one row to the routing sheet called latency-sensitive OpenAI jobs, then list only repeat tasks that currently feel slow or expensive.

T1OpenAI News08/25T1OpenAI News08/25

Small Money Systems

Small, repeatable, real

Use it

Small system: ChatGPT Work admin audit mini-retainer

System: build a monthly ChatGPT Work and Codex admin audit that checks usage, members, permissions, and limits. Customer: small production companies, agencies, or Japan-entry teams already experimenting with AI. Offer: a one-page risk and cleanup report plus a 30-minute walkthrough. Price: ¥45,000 per month. Existing assets: Grey Group OS, Goodsense operating notes, SET proposal structure, and Adrian’s own AI workspace habits. AI workflow: ChatGPT Work summarizes usage notes, Codex formats the report template, and Claude Code can turn it into a repeatable internal form. First action: draft the one-page audit template today. Repeatability: the same checklist runs every month. Effort: one hour. Expected value: recurring service revenue and a low-friction lead into larger workflow cleanup.

T1OpenAI News08/25
Test it

Small system: Copilot MCP setup pack for small dev teams

System: package a GitHub Copilot customization and MCP setup session. Customer: small software teams, creative studios with repos, or local businesses using GitHub but not structured AI workflows. Offer: a basic Copilot Customize tab setup, approved tool list, repo instructions, and handoff note. Price: ¥60,000 once, with optional monthly upkeep. Existing assets: Grey Group OS repo habits, Claude Code workflows, Codex review patterns, and Netlify deployment knowledge. AI workflow: Claude Code drafts repo instructions, Codex checks the setup against the repository, and ChatGPT prepares the client handoff. First action: make a sample setup checklist for one existing repo. Repeatability: each new client gets the same setup path with different tools. Effort: half day. Expected value: paid setup revenue and future support leads.

T1GitHub Changelog08/25

Build Next

Deployable now

Use it

Build a workspace admin snapshot

What to build: a one-page weekly snapshot that records ChatGPT Work and Codex members, permission changes, usage notes, and limit exceptions. Why now: OpenAI added an Admin plugin that brings those controls into the workspace. Effort: one hour. Expected impact: fewer access mistakes, clearer handoffs, and less time spent reconstructing who changed what. Dependencies: an active ChatGPT Work and Codex workspace, admin access, and one place to store the snapshot.

T1OpenAI News08/25
Test it

Build a Copilot customization starter file

What to build: a starter package for one GitHub repo that defines approved MCP tools, project knowledge, review rules, and model-use notes for Copilot. Why now: GitHub made the Copilot app Customize tab generally available with MCP support. Effort: half day. Expected impact: cleaner agent work and less repeated instruction writing. Dependencies: a GitHub repo, Copilot access, at least one MCP tool worth connecting, and a maintainer willing to test the workflow.

T1GitHub Changelog08/25

Try This Today

One action, right now

Open the workspace admin view and write down three things: who has access, what limits are active, and which Codex tasks ran this week. Save it as the first weekly admin snapshot. Do not optimize it yet.