Daily AI Zine
Thursday, September 24, 2026
Issue No. 069 · Tokyo · full edition

✦ Today's Big Thing

Copilot agents get a visibility layer

GitHub’s new OpenTelemetry support turns Copilot from a black-box helper into something teams can start measuring.

5 min read · 7 sections

In Brief
  1. GitHub Copilot now supports OpenTelemetry configuration through enterprise-managed settings.
  2. OpenAI’s frontier-model customer stories are converging on one practical theme: faster engineering work with clearer cost controls.
  3. Copilot for JetBrains adds agent approvals, stronger conversation controls, and shared organizational instructions.
  4. A new video-generation agent paper points to a useful creative pattern: generate, check, revise, then deliver.

Today's Big Thing

The one thing that matters

Test it

GitHub gives Copilot agents an observability layer

The practical shift is simple: agent work can now be watched like software operations, not just judged by the final diff.

GitHub Copilot now supports OpenTelemetry configuration through enterprise-managed settings. OpenTelemetry, often shortened to OTel, is an open-source way to collect traces, metrics, and logs from software systems. For agentic coding, that matters because the useful questions are no longer only “did the code pass?” They are also which model was used, which tools were called, where the agent stalled, and whether a workflow is getting cheaper or noisier over time. This is not a flashy model launch, but it is a serious operations feature. Treat it as the start of agent accountability inside GitHub, especially for teams running multiple coding agents against production repos.

My AI Ecosystem

Your actual stack

Watch it

Airbnb widens frontier-model access for engineering teams

OpenAI says Airbnb is expanding access to GPT-6 Astra and other frontier models to help engineering teams solve bugs, design systems, and ship faster. The useful pattern is not the logo. It is access control plus engineering workflow. Frontier models are moving from special-task tools into normal software delivery lanes.

Test it

Copilot for JetBrains adds tighter agent controls

GitHub Copilot for JetBrains 1.18.0 adds AI-assisted tool approvals, more control over agent conversations, shared skills and instructions for organizations, and plan review with the Codex CLI. This is a practical governance update: agent work becomes easier to approve, steer, and reuse across a team instead of staying inside one developer’s session.

Watch it

Hot signal: video-generation agents are learning to check their own work

VideoGen-Agent is a research system for video generation that uses external tools to augment prompts, generate clips, and verify results. The practical signal is the workflow, not the paper: video AI is moving toward generate, inspect, correct, and regenerate loops. For production work, that points to fewer one-shot prompts and more review pipelines.

CoWork Corner

Claude CoWork, day to day

CoWork is quiet, so revisit the project boundary

Claude CoWork is the workspace Adrian uses for persistent projects, operational records, and AI-assisted workflows. No meaningful CoWork-specific product change was found in today’s candidates. The relevant platform signal is Anthropic’s continued focus on containing Claude across products as agents become more capable. The useful workflow today is to pick one persistent project and make its boundary explicit: what the assistant may read, what it may edit, what tool access is allowed, and what requires human approval. That keeps project memory useful without letting one workspace become an uncontrolled operations hub.

Tier 4 · quiet day, honest fallback

GPT Desk

OpenAI, ChatGPT, Codex

Test it

Turn frontier-model access into one engineering escalation lane

OpenAI’s Airbnb story says GPT-6 Astra and frontier models are being used for bugs, design systems, and faster shipping. Adrian should care because Grey Group OS, Goodsense, and Daily AI Zine all have small software surfaces where hard bugs and workflow polish matter more than raw code volume. Action: choose one repo issue this week, write a short bug brief, then run the same task through Codex and ChatGPT with GPT-6 Astra as the escalation model.

Use it

Make OpenAI usage analytics answer one business question

OpenAI is positioning ChatGPT Work and Codex analytics as a way to connect AI usage and spend to business outcomes. Adrian should care because Grey Group OS needs proof that AI workflows save time or improve delivery, not just more prompts. Action: create one weekly scorecard for Goodsense or SET with three columns only: task type, AI tool used, and human time avoided or quality improved.

Small Money Systems

Small, repeatable, real

Test it

Small system: bilingual call-agent readiness audit

System: build a bilingual call-intake audit that tells a small business which calls, chats, WhatsApp messages, or web inquiries could be handled by an AI agent. Customer: Japan-based service businesses that miss leads or spend staff time on repeat questions. Offer: a short intake map, sample scripts, risk notes, and a first automation recommendation. Price: ¥35,000. Existing assets: Goodsense positioning, Grey Group OS templates, Japan-English workflow, and production-style interview habits. AI workflow: ChatGPT and the OpenAI API summarize call types, draft scripts, and classify escalation points. First action: make a 10-question intake form today. Repeatability: reuse the same audit template for each client. Effort: one hour. Expected value: small paid audits that can turn into implementation retainers.

Use it

Small system: AI usage value scorecard

System: build a one-page AI usage value scorecard for small teams already paying for ChatGPT Work, Codex, or similar tools. Customer: founders, agencies, and production teams that feel AI is useful but cannot explain the return. Offer: a spend-to-output review, training gaps, and three workflow fixes. Price: ¥50,000. Existing assets: Grey Group OS, Daily Ops, proposal templates, and Adrian’s own AI operating habits. AI workflow: ChatGPT summarizes usage notes, Codex helps generate the repeatable report template, and Claude Code can turn it into a small internal web form. First action: draft the scorecard fields. Repeatability: run it monthly as a lightweight retainer. Effort: half day. Expected value: savings, clearer renewals, and lead generation for deeper workflow work.

Build Next

Deployable now

Test it

Build an agent trace ledger for one repo

What to build: a small repo-level ledger that records each agent task, model choice, tool calls, review result, and whether the output shipped. Why now: GitHub Copilot now supports OpenTelemetry configuration through enterprise-managed settings, which makes agent behavior more measurable. Effort: half day. Expected impact: better review discipline and faster decisions about which agent workflows are worth keeping. Dependencies: access to Copilot enterprise-managed settings, one active GitHub repo, and a simple place to store the trace summary.

Try This Today

One action, right now

Pick one coding-agent task today and log four things: model used, tool approvals requested, time to useful output, and what the human reviewer changed. Do it manually first. If the pattern is useful, automate the ledger later.