Daily AI Zine
Sunday, August 16, 2026
Issue No. 031 · Tokyo · full edition

✦ Today's Big Thing

GitHub auth becomes agent infrastructure

No huge model launch today. The practical move is smaller and more useful: tighten the identity layer around the tools agents can touch.

4 min read · 7 sections

In Brief
  1. GitHub added safer OAuth and GitHub App options, including expiring access tokens and refresh tokens.
  2. Copilot for JetBrains now includes persistent memory, local model access, enterprise controls, and MCP reliability fixes.
  3. A new design-agent paper points toward assistants that remember reusable design process, not just final outputs.
  4. OpenAI’s Circles case is a useful template for turning Codex and API work into measurable retention and revenue workflows.

Today's Big Thing

The one thing that matters

Use it

GitHub makes app identity more agent-ready

A quiet day for major AI releases, but a useful security shift for anyone letting agents touch repositories and tools.

No stronger source-backed model or platform launch surfaced today. The useful shift is GitHub’s OAuth and GitHub App update: OAuth apps can opt in to expiring access tokens and refresh tokens, and GitHub has released multiple updates aimed at more secure app development. This matters because agent workflows increasingly depend on small connected apps rather than one big assistant. If an agent, connector, or internal tool keeps long-lived access to a repo, the risk is not only bad code, it is stale permission. Treat this as a prompt to audit which GitHub-connected tools still have broad or permanent access.

T1GitHub Changelog08/14

My AI Ecosystem

Your actual stack

Test it

Copilot for JetBrains gets local-model access

GitHub’s JetBrains update adds persistent memory, local model access through Ollama, more enterprise controls, and reliability fixes across MCP servers. The practical point is not model novelty. It is that coding assistants are becoming configurable workstations, with memory, local fallback, and admin policy in one place. Test it only if a JetBrains workflow is already active.

T1GitHub Changelog08/11
Watch it

Hot signal: design agents start remembering process

AutoDesign is a research item, not a product, but the direction is relevant: it treats design as a long task where the system can keep reusable experience from prior attempts. In plain terms, the assistant learns from the way a design was improved, not just the final file. Watch for this moving into deck, motion, and campaign tooling.

T1Hugging Face Daily Papers08/12
Watch it

Gemini API pushes managed agents toward remote MCP

Google says Gemini API Managed Agents added background tasks, remote MCP, and other capabilities for production-ready agents. Remote MCP means an agent can reach tools and data sources without everything living on the same machine. This is outside the core daily stack, but it is a useful signal for where connector architecture is heading.

T1Google AI Blog07/07

CoWork Corner

Claude CoWork, day to day

CoWork: write the tool’s exit rule first

Claude CoWork is the workspace Adrian uses for persistent projects, operational records, and AI-assisted workflows. No T1 source in today’s pool confirmed a meaningful new CoWork product change. The useful workflow to revisit is Claude Desktop’s one-click MCP server installation: easier installation makes it more important to write the tool’s exit rule before use. For any connector, add one sentence to the room note: what it may touch, what it must not touch, and when the session should stop using it.

T1Anthropic Engineering Blog

GPT Desk

OpenAI, ChatGPT, Codex

Test it

Use Codex to build one retention loop, not another demo

What changed: OpenAI’s Circles case says the company uses the OpenAI API and Codex for AI-native telco experiences, with reported gains in ARPU, churn, and development efficiency. Why Adrian cares: Goodsense and Grey Group OS need repeatable operating loops, not isolated prompts. What to do today: pick one existing lead, sponsor, or production contact list and have Codex draft a tiny retention workflow that scores recent activity, writes the follow-up reason, and produces the next email.

T1OpenAI News08/03

Small Money Systems

Small, repeatable, real

Use it

Small system: OAuth connector preflight mini-audit

System: build a lightweight OAuth and GitHub App preflight for small teams using agents or repo tools. Customer: Japan-based startups, agencies, and production-adjacent teams with GitHub-connected AI tools. Offer: a short report listing redirect URI, token lifetime, refresh-token, and overbroad-access issues. Price: ¥35,000 to ¥60,000 per audit. Existing assets: Grey Group OS, GitHub experience, AI workflow notes, and bilingual business documentation habits. AI workflow: Claude Code or Codex reviews exported app settings and drafts the findings. First action: make a one-page intake form today. Repeatability: the same checklist can run for every team before an AI rollout. Effort: one hour. Expected value: small audit revenue plus a lead-in to monthly AI operations support.

T1GitHub Changelog08/14

Build Next

Deployable now

Use it

Build a GitHub OAuth inventory note

What to build: a small repo note generator that lists connected OAuth apps and GitHub Apps, then marks which ones need expiring tokens, refresh tokens, or narrower redirect settings. Why now: GitHub’s latest app-platform update puts token expiry and refresh behavior back on the table. Effort: half day. Expected impact: fewer stale permissions and faster client-facing security checks. Dependencies: GitHub admin access and a manual export or screenshot of current app settings.

T1GitHub Changelog08/14

Try This Today

One action, right now

Open one active repository and list every connected app or OAuth integration with write access. Mark whether it has a clear owner, a narrow job, and a reason to keep access. Remove or flag one stale connection before doing any agent work.