Daily AI Zine
Sunday, August 2, 2026
Issue No. 018 · Tokyo · full edition

✦ Today's Big Thing

Codex is moving into creative tool work

The useful shift today is not a new model. It is OpenAI showing Codex as a way for creative teams to build their own working tools faster.

4 min read · 7 sections

In Brief
  1. Codex is being shown as a creative prototyping partner, not only a coding assistant.
  2. Claude Opus 5 is positioned for long-running agents, coding, and professional work.
  3. GitHub’s Copilot app points toward project spaces built around agents and canvases.
  4. A new agent-memory paper is worth watching for multi-agent reliability.

Today's Big Thing

The one thing that matters

Use it

Codex is becoming a creative operations tool

OpenAI says its creative team uses Codex to build custom creative tools, accelerate ideation, and prototype faster with context-aware AI. This is not a model launch. It is a useful shift in where coding agents belong: closer to decks, treatments, internal calculators, research helpers, and production utilities, not only software engineering tickets. For Adrian, the signal is practical. The next useful Codex experiment should start from a recurring creative bottleneck, then turn it into a small internal tool that saves one producer or designer from doing the same setup work again.

My AI Ecosystem

Your actual stack

Test it

Claude Opus 5 belongs in the long-running lane

Anthropic positions Claude Opus 5 as a step-change improvement for the Opus tier, powering long-running agents and improving coding and professional work. Treat it as the model for harder delegated runs, not the default for every prompt. The practical test is one contained multi-step coding or operations task with a written stop condition.

Test it

GitHub’s Copilot app points to agent project spaces

GitHub’s Copilot app guidance now centers on starting projects, working with AI agents, using canvases, and managing development workflow. The useful part is not beginner onboarding. It is the product shape: coding agents are being wrapped in project surfaces where planning, state, and edits sit together.

Watch it

Hot signal: agent teams need memory of who was reliable

Σ-Mem proposes an online reliability memory for multi-agent systems. In plain terms, it records which agent has been competent in which situation, instead of trusting every peer response equally. This is research, not a product to install today, but the direction matters for any future multi-agent operating stack.

CoWork Corner

Claude CoWork, day to day

CoWork: hold the workspace steady today

Claude CoWork is the workspace Adrian uses for persistent projects, operational records, and AI-assisted workflows. No meaningful CoWork product change was found in today’s candidates. The useful revisit is to treat CoWork as the written control layer for long-running agent work: task brief, source files, approval boundary, result summary, and next handoff. That matters more as Claude Opus 5 is positioned for long-running agents and professional work.

Tier 4 · quiet day, honest fallback

GPT Desk

OpenAI, ChatGPT, Codex

Use it

Use Codex to build one creative ops utility

OpenAI says its creative team uses Codex to build custom creative tools and prototype faster. Grey Group should treat that as permission to move Codex into production operations, not just repos. Pick one repeat annoyance inside Goodsense or SET, such as converting a rough brief into a structured treatment skeleton, and ask Codex to build a tiny local utility or Netlify preview that does only that one job.

Small Money Systems

Small, repeatable, real

Test it

Small system: paid AI repo kickoff canvas

System: Build a one-hour AI repo kickoff canvas for small teams starting a site, campaign tool, or internal dashboard. Customer: Japan-based founders, agencies, and production teams that need a working technical plan before hiring help. Offer: They receive a GitHub project structure, first task list, agent instructions, and a Netlify-ready deployment path. Price: ¥35,000. Existing assets: Grey Group OS, GitHub, Netlify, Claude Code, Codex, and Adrian’s production planning habits. AI workflow: Use the GitHub Copilot app pattern of projects, agents, and canvases, then let Claude Code or Codex draft the repo scaffold and task file. First action: Make one sample canvas for a fake landing-page build today. Repeatability: Reuse the same template for every paid setup. Effort: one hour. Expected value: small setup revenue and warmer implementation leads.

Build Next

Deployable now

Use it

Build a tiny creative-tool request form

What to build: A simple internal form that captures a recurring creative or production task, asks for inputs, expected output, owner, and deadline, then turns it into a Codex-ready build brief. Why now: OpenAI is explicitly showing Codex as useful for custom creative tools, not only software tickets. Effort: one hour. Expected impact: better briefs and fewer vague coding-agent runs. Dependencies: an existing GitHub repo, Codex access, and one small workflow worth turning into a utility.

Try This Today

One action, right now

Pick one repetitive creative task from this week. Write the inputs, output format, success check, and stop condition in ten lines. Then hand only that brief to Codex. Do not ask for a platform. Ask for one working utility.