Daily AI Zine · No. 048

Today's Big Thing

The one thing that matters

Test it

Claude Fable 5.1 changes the agent-cost test

The new Anthropic models matter less as a benchmark headline and more as a routing decision for coding, research, and long-running agent work.

Anthropic introduced Claude Fable 5.1 and Claude Mythos 5.1 as its most advanced models for coding and knowledge work. A second report says Fable 5.1 is typically around 25 percent cheaper than Fable 5, and up to 45 percent cheaper for complex agentic tasks. The practical move is not to swap every job to the new model. It is to rerun the jobs where model cost previously changed the plan: long coding sessions, research-heavy briefs, and agent runs with many tool calls. Verdict: test. Run one controlled comparison against the current default model before changing the stack.

T1Anthropic Newsroom T2The Verge09/01

My AI Ecosystem

Your actual stack

Use it

GitHub CLI makes visual review less manual

GitHub CLI now has a repeatable --attach flag that uploads a local image or video and references it inline in an issue, pull request, or comment. For product, design, and production sites, this removes the browser drag-and-drop step from visual QA. Verdict: use. Add it to bug reports, PR review notes, and release checklists where screenshots or clips prove the issue faster than text.

T1GitHub Changelog09/01

Use it

GitHub budgets now fit temporary work

GitHub now lets an individual user budget carry an optional expiration date, and removes that budget automatically on the date set. That matters for temporary contributors, short implementation sprints, and controlled AI-tool spending. Verdict: use. Treat expiring budgets as the default for time-boxed work instead of relying on someone to remember cleanup later.

T1GitHub Changelog09/01

Watch it

Hot signal: persistent generated worlds keep improving

Matrix-Game 3.5 is a research item on real-time streaming interactive world models, meaning AI-generated environments that respond as a user moves through them. The summary claims improvements in scene stability, geometry, camera control, and longer interaction through patch memory. Verdict: watch. This is not production tooling yet, but it points toward future previsualization, XR, and interactive pitch prototypes.

T1Hugging Face Daily Papers08/29

CoWork Corner

Claude CoWork, day to day

CoWork: prune the room before adding tools

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. The useful workflow to revisit is connector restraint: Claude Desktop has a documented one-click MCP server installation pattern, but each room should only carry the tools needed for that room’s current job. Before adding a connector, write the job it serves, the data it can touch, and the condition for removing it.

T1Anthropic Engineering Blog

GPT Desk

OpenAI, ChatGPT, Codex

Use it

Astra raises the security-briefing bar

OpenAI says Astra is the first OpenAI model to meet the Critical cybersecurity capability threshold under its Preparedness Framework, with stronger safeguards for release. Adrian should care because Grey Group OS, Goodsense, and SET all benefit from a plain-English rulebook for which AI tools may touch code, credentials, client research, and production documents. Action: write a one-page internal note today that separates allowed tasks, approval-needed tasks, and never-automated tasks for ChatGPT, Codex, and API use.

T1OpenAI News09/01

Test it

OpenAI’s workflow cases point to operating assets

OpenAI says Basis, Clay, and Exa Labs use AI agents to improve onboarding, account management, and developer integrations. Adrian should care because the same pattern can turn Daily Ops and Grey Group OS from notes into repeatable operating assets. Action: pick one live workflow, such as client onboarding or proposal follow-up, and have ChatGPT map the steps while Codex turns the repeatable part into a checklist, script, or tiny internal app.

T1OpenAI News09/01

Small Money Systems

Small, repeatable, real

Use it

Small system: launch asset cleanup pack

System: build a fixed-scope launch asset cleanup pack for rushed campaigns. Customer: small brands, agencies, and founders preparing a product, event, or content launch. Offer: they receive sharpened copy, a release checklist, a landing-page punch list, and five reusable announcement variants. Price: ¥55,000 per pack. Existing assets: Goodsense proposal instincts, Grey Group production judgement, and past deck workflows. AI workflow: ChatGPT drafts variants, Codex or Claude Code formats checklists and page QA notes, and the API can later standardize intake. First action: choose one old proposal and convert it into the intake form today. Repeatability: every launch uses the same intake, review, and delivery structure. Effort: one hour. Expected value: small repeat revenue and faster proposal closeout.

T1OpenAI News08/20

Test it

Small system: onboarding room setup

System: build a paid onboarding room setup for teams that are trying to make AI useful but still lose decisions in chat threads. Customer: Japan-English operators, small agencies, and founder-led teams. Offer: one structured project room, an account-management note template, a handoff checklist, and a weekly summary format. Price: ¥75,000 for setup. Existing assets: Grey Group OS, Daily Ops, Claude CoWork habits, and commercial production handoff experience. AI workflow: ChatGPT maps the workflow, Claude Code or Codex turns it into reusable markdown templates, and CoWork holds the persistent record. First action: package one existing internal room as a sanitized sample. Repeatability: the same room pattern sells to each new team with light edits. Effort: half day. Expected value: setup revenue plus retainer leads.

T1OpenAI News09/01

Build Next

Deployable now

Use it

Build a visual QA commenter for GitHub

What to build: a tiny script that takes a screenshot or short screen recording, opens a GitHub issue or PR comment, and attaches the file through the new GitHub CLI media flag. Why now: GitHub CLI now supports repeatable media attachments directly in issues, pull requests, and comments. Effort: one hour. Expected impact: faster visual QA and clearer handoff on site, deck, and app fixes. Dependencies: GitHub CLI access, a target repository, and a local folder where review images or clips are saved.

T1GitHub Changelog09/01

Use it

Build a temporary GitHub budget check

What to build: a small admin checklist or script that records who has a temporary GitHub user budget, why it exists, and when it expires. Why now: GitHub now supports expiration dates on individual user budgets and removes them automatically on that date. Effort: 15 minutes. Expected impact: lower tool-spend leakage and cleaner short-term collaborator governance. Dependencies: GitHub admin access and an existing list of users who receive temporary budgets.

T1GitHub Changelog09/01

Try This Today

One action, right now

Pick one open GitHub issue or PR that would be clearer with a screenshot or short clip. Use the GitHub CLI media attachment path, then add one sentence explaining what the image proves. Make visual QA a habit, not a browser chore.