Daily AI Zine
Thursday, September 17, 2026
Issue No. 062 · Tokyo · full edition

✦ Today's Big Thing

OpenAI pushes agents into advertising

Today’s useful shift is not another model race. It is AI moving closer to paid customer journeys, measurement, and repeatable commercial systems.

5 min read · 7 sections

In Brief
  1. OpenAI introduced AI advertising experiences including Sponsored Agents, marketer tools, and HubSpot and Shopify integrations.
  2. ChatGPT Work and Codex analytics now have a clearer business-value frame for usage, spend, and training gaps.
  3. GitHub added VS Code Agents activity to Copilot usage metrics, making agent adoption easier to measure.
  4. Claude CoWork and chat are reportedly being merged into one Claude, with Docs and Slides also reported as new creation tools.
  5. A new design-agent paper points to a practical pattern: choose the design direction before generating frontend code.

Today's Big Thing

The one thing that matters

Test it

OpenAI turns ads into agent workflows

The day’s clearest development is commercial, not technical.

OpenAI introduced AI advertising experiences that include Sponsored Agents, tools for marketers, and integrations with HubSpot and Shopify. The practical change is that ads can start looking less like static placements and more like guided customer interactions, where an agent can explain, qualify, and route a prospect. For a production and brand operator, this is worth testing before it becomes a default agency pitch format. Treat it as a controlled sales experiment, not a media-buy assumption: pick one real offer, script the customer path, define the handoff point, and compare the result against a normal landing page or deck follow-up.

T1OpenAI News09/16

My AI Ecosystem

Your actual stack

Use it

ChatGPT Work and Codex get value accounting

OpenAI is framing ChatGPT Work and Codex analytics around usage, spend, training needs, and business outcomes. This matters because agent work is now a management problem as much as a tool problem. The useful move is to stop counting prompts and start counting which workflows produced usable output.

T1OpenAI News09/16
Use it

GitHub adds VS Code Agents to Copilot metrics

GitHub Copilot usage metrics now include generally available metrics for activity in the dedicated VS Code Agents window. For teams using coding agents, this turns a previously fuzzy behavior into something managers can review: adoption, engagement, and whether agent work is spreading beyond one power user.

T1GitHub Changelog09/11
Watch it

Hot signal: design agents separate direction from code

A new paper on vibe design agents describes a useful pattern: let the AI choose and expose the design direction before it writes the interface code. In plain terms, ask for three coherent creative routes first, then build one. That is more useful than asking for one polished page and hoping the taste is right.

T1Hugging Face Daily Papers09/13

CoWork Corner

Claude CoWork, day to day

CoWork and chat start converging

Claude CoWork is the workspace Adrian uses for persistent projects, operational records, and AI-assisted workflows. Claude CoWork and regular Claude chat are reported to be merging into one Claude, with rollout to Pro and Max plans first across web, desktop, and mobile over the coming weeks. A separate report says Claude is also adding Docs and Slides for creating, exporting, editing, and sharing documents and presentations from chats. Treat this as a workspace-boundary change: before moving live work, keep one clear rule for what belongs in persistent project memory versus a quick disposable chat.

T2Simon Willison Weblog09/16T2The Verge09/16T1Anthropic Newsroom

GPT Desk

OpenAI, ChatGPT, Codex

Test it

Sponsored Agents deserve one real sales test

OpenAI introduced AI advertising experiences including Sponsored Agents, marketer tools, and HubSpot and Shopify integrations. Adrian should care because Goodsense and Street Attack Japan both benefit from pitches that show an interactive buyer path, not just a visual concept. Do this today: write one Sponsored Agent flow for a real offer, with greeting, qualification, objection answer, and handoff, then turn it into a one-page sales demo.

T1OpenAI News09/16
Use it

Make OpenAI analytics part of Grey Group OS

OpenAI is pointing ChatGPT Work and Codex analytics toward usage, spend, training needs, and business outcomes. Adrian should care because Grey Group OS needs to prove which AI workflows actually save time or improve delivery. Do this today: create one analytics note with three columns, workflow, AI cost or usage signal, and output accepted, then fill it for one Codex task and one ChatGPT task.

T1OpenAI News09/16
Use it

Turn misalignment reports into a model oddity log

OpenAI published a framework for tracking, investigating, and disclosing model misalignment, alongside six reports of unexpected or concerning model behavior. Adrian should care because Codex and ChatGPT outputs can enter Grey Group OS, proposals, and production planning if no one flags odd behavior. Do this today: add a model oddity log with date, tool, prompt, strange behavior, business risk, and whether the output was used.

T1OpenAI News09/16

Small Money Systems

Small, repeatable, real

Test it

Small system: Sponsored Agent mockup mini-pack

System: build a one-page Sponsored Agent concept pack for a brand offer. Customer: Tokyo agencies, local retailers, DTC brands, or tourism-adjacent operators that want to see how interactive AI ads might sell. Offer: they receive a customer journey, three conversation scripts, a basic handoff map for HubSpot or Shopify, and a simple demo page. Price: ¥55,000 per pack. Existing assets: Goodsense positioning, Street Attack Japan activation logic, proposal templates, and Adrian’s production sense. AI workflow: ChatGPT drafts the buyer paths, Codex or Claude Code turns the best path into a Netlify-ready demo page. First action: pick one past pitch and rewrite it as a customer-agent conversation. Repeatability: the same template works per sector. Effort: half day. Expected value: paid discovery work and warmer leads for larger production or campaign proposals.

T1OpenAI News09/16
Test it

Small system: AI spend-to-output tune-up

System: sell a compact AI usage review that connects spend and activity to accepted output. Customer: small companies already paying for ChatGPT, Codex, or GitHub Copilot but unsure what value they get. Offer: they receive a two-page report showing active workflows, waste, training gaps, and three changes to improve output. Price: ¥80,000 for the first review, then ¥50,000 monthly if continued. Existing assets: Grey Group OS, Daily Ops habits, Codex logs, and GitHub workflow experience. AI workflow: ChatGPT summarizes usage notes, Codex helps normalize task logs, and GitHub metrics support coding-agent activity where available. First action: draft a five-question intake form. Repeatability: the same report format runs monthly. Effort: half day. Expected value: small recurring retainer revenue and clearer savings for the client.

T1OpenAI News09/16T1GitHub Changelog09/11

Build Next

Deployable now

Test it

Build an AI ad concept microsite generator

What to build: a Netlify microsite generator that turns a product brief into an interactive ad concept with headline, scripted agent conversation, CTA, and handoff notes. Why now: OpenAI has moved advertising into agent-style experiences with Sponsored Agents and marketer integrations. Effort: half day. Expected impact: faster pitch prototypes and better client understanding of interactive ad formats. Dependencies: an existing Netlify account, a simple template repo, and access to ChatGPT or the OpenAI API.

T1OpenAI News09/16
Use it

Build an agent value dashboard

What to build: a small dashboard that combines ChatGPT Work, Codex, and GitHub agent activity into one weekly value view: tasks attempted, outputs accepted, costs or usage signals, and training gaps. Why now: OpenAI is emphasizing business-value analytics, and GitHub now includes VS Code Agents in Copilot usage metrics. Effort: one hour for a manual version. Expected impact: faster decisions on which agent workflows deserve more use. Dependencies: access to the relevant analytics exports or screenshots.

T1OpenAI News09/16T1GitHub Changelog09/11

Try This Today

One action, right now

Pick one offer and write the four-part Sponsored Agent path: greeting, qualification, objection answer, and handoff. Keep it to one page. If it reads like a useful salesperson, turn it into a Netlify mockup next.