Daily AI Zine
Saturday, August 29, 2026
Issue No. 044 · Tokyo · full edition

✦ Today's Big Thing

OpenAI’s Cursor decision is a dependency warning

The day’s most useful signal is not another model claim. It is a reminder to keep AI tooling replaceable.

5 min read · 7 sections

In Brief
  1. OpenAI says it will wind down its model contract with Cursor after Cursor’s acquisition by SpaceX.
  2. GitHub Copilot’s August Visual Studio update adds more control over reasoning, model choice, shared agents, and review timing.
  3. Magpie is an early research signal for real-time generated game worlds, useful to watch for creative prototyping.
  4. Cloudflare BotBase gives bot operators a dashboard for submissions, status edits, and behavior declarations.

Today's Big Thing

The one thing that matters

Use it

OpenAI’s Cursor cutoff turns model access into an operating risk

Platform dependency is now part of AI tool selection.

OpenAI says it will wind down its contract providing OpenAI models to Cursor after Cursor’s acquisition by SpaceX. The practical read is simple: AI coding tools are no longer only feature choices, they are supply-chain choices. A tool can be excellent and still become risky if its model access changes because of ownership, partnership, or platform conflict. For Adrian, the lesson is not to avoid Cursor specifically. It is to keep coding workflows portable across Claude Code, Codex, GitHub, and local repo conventions. Any live project that depends on one assistant, one model, or one hosted workspace needs a short escape plan.

T1OpenAI News08/28

My AI Ecosystem

Your actual stack

Watch it

GitHub Copilot adds more control in Visual Studio

GitHub’s August Visual Studio update says Copilot now gives users more control over how it reasons, which models are used, how teams share specialized agents, and when to request code review. This matters less as a Visual Studio-only item and more as a direction: agentic coding tools are moving from chat helpers toward governed delivery surfaces.

T1GitHub Changelog08/28
Test it

GitHub keeps pushing Copilot into shared work sessions

GitHub’s weekly Copilot release note points to more control over how Copilot runs, including team sessions in Slack and Teams plus customization across the app, CLI, and IDE. The useful pattern is shared agent work with visible context, not private one-off prompting. Test only if the team already has a repo workflow worth formalizing.

T1GitHub Changelog08/28
Watch it

Hot signal: Magpie points at real-time generated game worlds

Magpie is a research item on real-time world rendering for interactive games. The summary frames the core problem clearly: game prototypes need stable, reproducible visuals that respond to play, not just linear video generation. This is still research, but it is relevant to creative production because the line between video model, previz tool, and interactive pitch prototype is getting thinner.

T1Hugging Face Daily Papers08/26
Watch it

Cloudflare gives bot operators a clearer listing path

Cloudflare’s BotBase update adds a dashboard home for bot operators, with submission status tracking, submission editing, and a behavior model for declaring how bots use content. The practical change is governance language for agents that browse or consume websites. Watch this if public content, search visibility, or agent access becomes part of a client delivery.

T1Cloudflare Blog08/28

CoWork Corner

Claude CoWork, day to day

CoWork: turn repeat judgment calls into skill cards

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, so use the existing workspace discipline. Anthropic’s Agent Skills framing is a useful reminder: when an agent performs a recurring workflow, package the instruction as a reusable skill, meaning a compact playbook the agent can apply again. Start with one judgment call that repeats often, such as whether a brief is ready to quote, approve, or send back.

T1Anthropic Engineering Blog

GPT Desk

OpenAI, ChatGPT, Codex

Use it

Audit OpenAI dependencies outside the core stack

OpenAI says it will wind down its model contract with Cursor after the SpaceX acquisition. Adrian should care because Goodsense, SET, Street Attack Japan, and Grey Group OS should not depend on any one AI coding vendor’s commercial relationship. Today’s action: make a short list of repos, workflows, and proposal assets that assume a specific assistant or hosted coding tool, then mark the fallback as Codex, Claude Code, or manual GitHub review.

T1OpenAI News08/28
Test it

Give ChatGPT Work one campaign packaging trial

OpenAI says Stampli used Codex and ChatGPT Work to cut launch hours by 68% when a deadline was fixed and design resources were already committed elsewhere. Adrian should care because that is close to real production pressure: proposal assets, decks, treatments, recaps, and bilingual follow-ups. Today’s action: run one Goodsense or SET campaign package through ChatGPT Work as a single room with inputs, asset gaps, owners, and final send checklist.

T1OpenAI News08/20

Small Money Systems

Small, repeatable, real

Test it

Small system: campaign packaging room

System: build a fixed-deadline campaign packaging room that turns raw notes, visuals, and client asks into a sendable proposal pack. Customer: small brands, agencies, or event teams that need fast presentation polish. Offer: one cleaned deck outline, asset gap list, English-Japanese message pass, and final send checklist. Price: ¥60,000 per package. Existing assets: Goodsense proposal habits, SET production planning, Grey Group OS templates, and Adrian’s bilingual workflow. AI workflow: ChatGPT Work organizes the room while Codex or Claude Code generates repeatable checklist and file-structure helpers. First action: choose one recent messy proposal and rebuild it as the sample pack today. Repeatability: each room becomes the template for the next client. Effort: half day. Expected value: paid proposal support and faster sales turnaround.

T1OpenAI News08/20
Use it

Small system: team-session starter pack

System: create a setup pack for teams that want AI coding or content sessions without losing control of context. Customer: small dev teams, production tech vendors, or creative studios already using GitHub. Offer: one shared session brief, agent rules, review timing guide, and customization starter file. Price: ¥45,000 for a one-hour setup plus written handoff. Existing assets: Grey Group OS operating rules, Claude Code habits, Codex experiments, and GitHub repo practice. AI workflow: Claude Code drafts the files, Codex checks repo fit, and ChatGPT turns it into client-facing language. First action: make the starter file from one live repo. Repeatability: the same pack can be resold with light repo-specific edits. Effort: one hour. Expected value: small consulting revenue and fewer chaotic agent sessions.

T1GitHub Changelog08/28

Build Next

Deployable now

Use it

Build an AI vendor dependency manifest

What to build: a small repo file and script that lists which AI tools, models, APIs, and hosted assistants a project assumes. Why now: OpenAI’s Cursor decision shows that model access can change for business reasons, not only technical ones. Effort: one hour. Expected impact: less delivery risk when a tool changes access, pricing, or ownership. Dependencies: at least one active GitHub repo and a current list of AI tools used in that project.

T1OpenAI News08/28

Try This Today

One action, right now

Open one active repo or project folder. Add a short note with three lines: primary AI tool, fallback AI tool, and what must stay human-reviewed. Keep it boring, visible, and easy to copy into the next project.