Daily AI Zine · No. 008

Today's Big Thing

The one thing that matters

Test it

GitHub Copilot adds Gemini 3.6 Flash

The practical shift is not model bragging. It is that Copilot users now get another current model option inside the place coding work already happens.

GitHub says Gemini 3.6 Flash, Google’s latest Flash model, is rolling out in GitHub Copilot. The summary says it is designed for web and app development, coding, and longer-horizon agentic tasks, with configurable settings. Treat this as a useful model-choice moment: the same coding task can now be compared inside the GitHub workflow instead of moved between separate tools. For daily shipping, the action is simple: pick one real app task, run it through the usual coding agent path and through Copilot with Gemini 3.6 Flash, then compare correctness, patch size, and review burden. Verdict: test, not switch by default.

T1GitHub Changelog07/21

My AI Ecosystem

Your actual stack

Use it

GitHub Code Quality is now GA

GitHub Code Quality is now generally available on GitHub Enterprise Cloud and GitHub Team. GitHub frames the need clearly: AI accelerates code output, and teams need a quality layer that catches problems before the repo fills with fast but weak changes. Use it as a review backstop for AI-written pull requests, not as a replacement for human judgment.

T1GitHub Changelog07/20

Test it

OpenAI points ChatGPT Work at small business

OpenAI launched the ChatGPT for Small Businesses program. The stated focus is helping entrepreneurs build AI skills, automate work, and grow with ChatGPT Work. This matters because small-company AI adoption is moving from one-off prompting toward repeatable operating routines. Test it as a packaging signal: buyers may now understand ChatGPT Work as a business system, not just a chat window.

T1OpenAI News07/21

Watch it

Hot signal: real-time multimodal apps get deployment help

FlashRT is a research release for guiding agents that deploy real-time multimodal applications, including voice agents and interactive video generation. In plain English, it targets the hard part after a demo works: deciding how models, streaming, and parallel work should be arranged so the app responds quickly. This is not a product to rely on today, but it is a strong signal for voice, video, and live production tools.

T1Hugging Face Daily Papers07/19

CoWork Corner

Claude CoWork, day to day

Claude Desktop MCP remains the CoWork bridge

Claude CoWork is the workspace Adrian uses for persistent projects, operational records, and AI-assisted workflows. No fresh CoWork-specific product change showed up today. The useful adjacent capability to revisit is Claude Desktop Extensions, which Anthropic describes as one-click MCP server installation for Claude Desktop. MCP means Model Context Protocol, the connector layer that lets Claude reach approved tools and data sources. The practical move is to keep CoWork projects tied to only a small set of trusted connectors, rather than letting every project accumulate its own tool sprawl.

T1Anthropic Engineering Blog07/22

GPT Desk

OpenAI, ChatGPT, Codex

Use it

Turn ChatGPT Work into a client-facing setup sprint

OpenAI launched a ChatGPT for Small Businesses program focused on AI skills, automation, and growth with ChatGPT Work. Adrian should care because Goodsense and Grey Group OS already need repeatable operating systems, not more isolated prompts. Do this today: make a one-page ChatGPT Work setup menu with three automations, inquiry triage, proposal drafting, and meeting follow-up, then test it internally before selling it as a paid setup sprint.

T1OpenAI News07/21

Use it

Use the evaluation incident as a Codex safety check

OpenAI and Hugging Face shared early findings from a security incident during model evaluation, highlighting advanced cyber capabilities and lessons for defenders. Adrian should care because Codex work inside Grey Group OS, Goodsense, and production tooling can touch files, credentials, and deploy paths. Do this today: create a Codex run checklist that requires sandboxing, repo backups, and no full-access mode on client or production folders.

T1OpenAI News07/21

Small Money Systems

Small, repeatable, real

Use it

Sell a small-business ChatGPT Work setup sprint

System: build a fixed setup sprint that turns ChatGPT Work into three working routines for a small company. Customer: Tokyo founders, agencies, and production vendors who know they should use AI but have no operating method. Offer: a short setup, three reusable prompt workflows, and a one-page staff guide. Price: ¥75,000 for the first setup. Existing assets: Grey Group OS, Goodsense proposal habits, and Adrian’s own ChatGPT workflows. AI workflow: ChatGPT drafts the routines, Codex or Claude Code turns repeatable pieces into templates, and Adrian reviews the client fit. First action: write the three-workflow menu this hour. Repeatability: every client gets the same base system with light customization. Effort: one hour. Expected value: paid setup revenue plus lead generation for deeper AI workflow retainers.

T1OpenAI News07/21

Build Next

Deployable now

Test it

Build a model-lane smoke test for coding agents

What to build: a small GitHub repo with one real bug fix, one UI change, and one docs update, then run the same tasks through Claude Code, Codex, and GitHub Copilot with Gemini 3.6 Flash. Why now: Gemini 3.6 Flash is rolling into Copilot for web, app, coding, and longer-horizon agent tasks, so model choice can be tested inside the coding workflow. Effort: one hour. Expected impact: better tool selection and less wasted review time. Dependencies: active access to the three coding tools and a non-sensitive repo.

T1GitHub Changelog07/21

Try This Today

One action, right now

Pick one small repo task today. Run it through Claude Code, Codex, and GitHub Copilot with Gemini 3.6 Flash. Score only three things: did it work, how much review did it need, and would you trust it again for that task type.