Daily AI Zine · No. 016

Today's Big Thing

The one thing that matters

Use it

GitHub makes Copilot model defaults a policy decision

New models can become available by default, which shifts the work from setup to governance.

GitHub is introducing a global default enablement policy for generally available Copilot models on Copilot Business and Copilot Enterprise. Instead of requiring admins to manually enable each new model, GitHub is moving toward default availability for models that have reached general availability. The practical change is simple: model access can expand faster than the team’s written rules. For an AI-heavy operator, this means Copilot needs a model-routing note, not just an admin setting. Define which model lane is allowed for small edits, multi-file refactors, code review, and sensitive repository work. Verdict: use. Treat this as a governance update and write the routing rule before the next model appears in the tool picker.

T1GitHub Changelog07/29

My AI Ecosystem

Your actual stack

Use it

Copilot code review gets MCP and agent skills in general availability

GitHub says Copilot code review support for agent skills and MCP servers is now generally available for Copilot Pro, Pro+, Business, and Enterprise users. MCP is the connector standard that lets an AI tool reach approved external systems. This matters because code review can now pull in project-specific checks instead of only reading the diff. Verdict: use for one low-risk repo first.

T1GitHub Changelog07/29

Test it

OpenAI positions GPT-5.6 as an efficiency upgrade

OpenAI says GPT-5.6 improves efficiency across models, inference, and agentic workflows, with the stated aim of delivering more useful intelligence per dollar. The useful read is cost discipline: if a stronger model also reduces retries, it may be cheaper on real tasks even when the headline model looks expensive. Verdict: test with repeatable tasks, not impressions.

T1OpenAI News07/29

Watch it

Hot signal: video models are moving toward navigable worlds

Wonder is a research project for real-time, camera-controllable video world exploration. Given an image or a conditional video, it aims to create a playable scene where the user can move the camera and revisit observed areas. People are paying attention because this points beyond clip generation toward interactive previsualization. Verdict: watch. It is not a production tool yet.

T1Hugging Face Daily Papers07/27

CoWork Corner

Claude CoWork, day to day

CoWork should keep MCP installs boring

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 Claude Desktop’s documented one-click MCP server installation pattern: treat every connector as an operational dependency, not a novelty. Keep a short note beside each project that says what the connector can touch, why it is installed, and when it should be removed.

T1Anthropic Engineering Blog

GPT Desk

OpenAI, ChatGPT, Codex

Test it

Use GPT-5.6 as a Codex cost test, not a belief test

What changed: OpenAI says GPT-5.6 improves efficiency across models, inference, and agentic workflows. Why Adrian should care: Grey Group OS, Daily Ops, Netlify fixes, proposal utilities, and small internal tools all suffer when an AI run needs three retries. What to do: pick one recent Codex task, rerun it with a fixed prompt and success checklist, then compare completion quality and retry count against the previous model lane. Verdict: test.

T1OpenAI News07/29

Use it

Turn OpenAI’s researcher push into a proposal research template

What changed: OpenAI is giving 100,000 academic researchers free access to ChatGPT’s most advanced AI models for research, collaboration, and discovery. Why Adrian should care: Goodsense, SET, and Grey Group often need credible Japan-English briefing work before a pitch becomes a paid engagement. What to do: create one ChatGPT research template for market background, source questions, interview prep, and risk notes, then run it on the next proposal. Verdict: use.

T1OpenAI News07/29

Small Money Systems

Small, repeatable, real

Test it

Sell a small MCP code-review wiring pack

System: build a one-repo Copilot code-review setup that adds a simple agent skill and an MCP-connected project checklist. Customer: small studios, local SaaS teams, or production-adjacent companies with GitHub repos but weak review discipline. Offer: they receive one configured review lane, a short model-use note, and a handoff checklist. Price: ¥55,000 for the first repo. Existing assets: Grey Group OS review patterns, Claude Code, Codex, GitHub, and Adrian’s production checklist habits. AI workflow: Claude Code drafts the repo instructions, Codex tests one pull request path, and Copilot code review runs with MCP context. First action: choose one internal repo and write the review checklist today. Repeatability: the same setup becomes a fixed package for each new repo. Effort: half day. Expected value: small service revenue plus fewer review misses on internal builds.

T1GitHub Changelog07/29

Build Next

Deployable now

Use it

Build a Copilot model-lane file for active repos

What to build: a small repo file or internal page that lists which Copilot model lane to use for tiny edits, multi-file refactors, review, and off-limits work. Why now: GitHub is introducing default enablement for generally available Copilot models, so model access can change without one-off admin setup. Effort: one hour. Expected impact: fewer wrong-model runs and clearer review expectations. Dependencies: Copilot Business or Enterprise access, active GitHub repos, and one owner willing to enforce the note.

T1GitHub Changelog07/29

Try This Today

One action, right now

Open one active repo and add four lines: default model for small edits, model for refactors, model for review, and tasks that need human approval first. The point is not perfection. It is preventing Copilot’s model menu from becoming invisible policy.