Daily AI Zine · No. 075

Today's Big Thing

The one thing that matters

Test it

OpenAI turns DevDay into an agent platform reset

The important shift is not one model name. It is the amount of product surface now aimed at agents, Codex, shared workspaces, speed tiers, and developer distribution.

OpenAI’s DevDay recap says it announced more than 20 updates across GPT-6 Astra, ChatGPT, Codex, APIs, security, and new builder tools. Supporting coverage describes always-on agents called Dots, cloud execution for Codex, faster paid Astra access, and a team collaboration space inside ChatGPT. The practical read: OpenAI is trying to make ChatGPT and Codex less like separate tools and more like one operating layer for people, agents, and apps. Do not migrate workflows on announcement day. Pick one contained build task, one research task, and one collaboration task, then test whether the new OpenAI stack reduces handoff friction compared with the current setup.

T1OpenAI News09/29 T2ITmedia AI+09/29 T2Perplexity Research09/29

My AI Ecosystem

Your actual stack

Test it

GPT-6.1 Sol arrives in GitHub Copilot

GitHub says GPT-6.1 Sol, OpenAI’s latest model, is generally available and rolling out in Copilot. The stated use case is agentic coding and terminal workflows, which means it belongs in a controlled repo trial rather than a blanket default change. Give it one multistep issue with tests and compare the diff against the current preferred coding model.

T1GitHub Changelog09/29

Use it

GitHub repos can now carry outside business context

GitHub now lets external systems bring business context into repositories using external custom properties. That matters for AI work because agents need to know which repos are client-facing, internal, archived, production-critical, or experimental before changing anything. Start with read-only labels before using them to drive automation.

T1GitHub Changelog09/29

Test it

Hot signal: Runway pushes Gen-4.5 on motion and fidelity

Runway’s research index is positioning Gen-4.5 as a video model focused on motion quality, prompt adherence, and visual fidelity. The source is thin, so treat this as a creative-tool signal, not a production decision. It is worth testing for short previsualization, mood frames in motion, and pitch-deck inserts where perfect edit control is not required.

T1Runway Research

CoWork Corner

Claude CoWork, day to day

CoWork: local-task signals point back to Claude Code

Claude CoWork is the workspace Adrian uses for persistent projects, operational records, and AI-assisted workflows. No meaningful Tier 1 CoWork product change surfaced in today’s candidates. A user report says local computer tasks are being deprecated from October 6 and points new local work toward Claude Code, so treat it as a migration warning until Anthropic documents it. Today’s useful workflow is simple: list any CoWork task that depends on local files, then write one Claude Code handoff note with the goal, files, approval boundary, and stop condition.

Tier 4 · quiet day, honest fallback

T1Anthropic Engineering Blog T2r/ClaudeAI09/29

GPT Desk

OpenAI, ChatGPT, Codex

Test it

Use Codex cloud as a controlled Grey Group OS runner

What changed: OpenAI’s DevDay recap includes Codex among the major builder updates, and supporting material describes running Codex on demand locally, remotely, or in the cloud with reusable development environments. Why Adrian should care: Grey Group OS needs repeatable implementation work without exposing every repo to broad agent permissions. What to do: choose one low-risk GitHub issue, define the approved environment and permissions, run Codex in the cloud, then review the diff before merge.

T1OpenAI News09/29 T2Perplexity Research09/29

Test it

Treat Dots as follow-up drafts, not autonomy

What changed: OpenAI announced Dots, described in coverage as always-on assistants that can work across connected apps in the background and learn preferences over time. Why Adrian should care: Goodsense, SET, and Street Attack Japan all depend on follow-up discipline after calls, shoots, proposals, and introductions. What to do: test one Dots-style workflow only for draft reminders and next-step summaries. Do not let it send, book, or change records until the review loop is proven.

T1OpenAI News09/29 T2The Verge09/29

Test it

Astra Ultrafast is a latency test, not a default plan

What changed: OpenAI’s DevDay coverage describes an Ultrafast tier for GPT-6 Astra, with supporting Japanese coverage saying it can run up to eight times faster in some paid contexts. Why Adrian should care: faster responses matter for live research, client-facing demos, and Codex flow, but only if they change delivery speed. What to do: run one timed comparison on a proposal rewrite, a code task, and a bilingual research brief before paying for higher speed.

T1OpenAI News09/29 T1OpenAI News09/23 T2Perplexity Research09/29

Small Money Systems

Small, repeatable, real

Use it

Small system: Codex cloud fix sprint

System is a one-day website and repo cleanup sprint using cloud-run Codex for bounded fixes. Customer is a Japan-based small business, creator, or production vendor with a stale GitHub or Netlify site. Offer is three small fixes, a before-and-after note, and a plain-language risk report. Price is ¥35,000. Existing assets are Grey Group OS, GitHub habits, Netlify deployment experience, and Adrian’s proposal templates. AI workflow is Codex cloud for implementation, Claude Code for review, and ChatGPT for the client summary. First action is to list five friendly prospects with visible site issues today. Repeatability comes from reusing the same intake, issue template, and review checklist. Effort is half day. Expected value is small paid cleanup revenue and a path into retainers.

T1OpenAI News09/29 T2Perplexity Research09/29

Use it

Small system: GitHub repo context audit

System is a lightweight audit that adds business context to a client’s GitHub repositories before they let AI agents touch code. Customer is a small software team, studio, or operations-heavy company starting to use Copilot or Codex. Offer is a repo map with labels for production risk, ownership, client impact, and automation readiness. Price is ¥45,000. Existing assets are Grey Group OS operating discipline, GitHub experience, and AI workflow review patterns. AI workflow is Claude Code or Codex to inspect repo metadata, then ChatGPT to draft the management summary. First action is to create a one-page sample using Adrian’s own repo list. Repeatability comes from the same property schema and scoring rubric. Effort is half day. Expected value is paid advisory work and safer agent adoption conversations.

T1GitHub Changelog09/29

Build Next

Deployable now

Use it

Build a repo business-context map

What to build is a small GitHub repo dashboard that mirrors external custom properties into a human-readable table: owner, client impact, production risk, agent permission level, and last review date. Why now is GitHub’s new support for bringing outside business context into repositories. Effort is half day. Expected impact is fewer unsafe agent tasks and faster repo triage before coding work starts. Dependencies are a GitHub account, a short property schema, and enough repo metadata to classify the first batch.

T1GitHub Changelog09/29

Try This Today

One action, right now

Pick one low-risk GitHub issue today. Run it once with the current Copilot model and once with GPT-6.1 Sol. Compare time, tests, diff size, and review effort. Keep the better result, but save both notes for routing decisions.