Daily AI Zine · No. 009

Today's Big Thing

The one thing that matters

Test it

OpenAI Presence makes enterprise agents the main story

The new platform matters because it packages voice and chat agents for real customer and internal workflows.

OpenAI introduced Presence, an enterprise AI agent platform for deploying trusted voice and chat agents into customer and internal workflows. The important part is packaging. Instead of treating agents as custom demos, OpenAI is framing them as operational infrastructure for conversations, support, lead handling, and internal process work. For Adrian, the useful read is simple: voice and chat agents are moving from “can we build this?” to “which repeatable workflow deserves an agent?” The near-term move is not to replace teams. It is to identify one narrow, measurable handoff, such as inquiry triage, follow-up, briefing, or internal request routing, and test whether an agent reduces delay without losing quality.

T1OpenAI News07/22

My AI Ecosystem

Your actual stack

Test it

Copilot cost control becomes a real workflow decision

GitHub says Copilot now bills usage at listed API rates, which makes the old “Copilot versus raw API” question more practical. The value is no longer only model access. It is the surrounding coding workflow, policy, and harness work. For teams using multiple coding agents, this is a cue to compare tasks by lane: direct API for repeatable scripts, Copilot for repo work, and stronger agents only when the task needs it.

T1GitHub Blog07/22

Watch it

Hot signal: AI agents enter branded commercial production

Luma’s news index says South African agency Boundless delivered Mazda’s first AI-produced commercial using Luma Agents. The summary is thin, so the safe takeaway is narrow: AI creative agents are now being presented as part of commercial campaign delivery, not only mood-board generation. Treat this as a signal to test production-adjacent use, such as concept variants, previs, recap edits, and sponsor mockups, not final-client replacement.

T1Luma AI News07/23

Test it

Runway pushes Gen-4.5 as a higher-control video model

Runway describes Gen-4.5 as its best video model, with state-of-the-art motion quality, prompt adherence, and visual fidelity. The practical question is whether it reduces the number of failed generations for production planning assets. Test it only on controlled deliverables: one reference frame, one camera move, one short prompt, and a pass-or-fail judgment against the brief.

T1Runway Research07/23

CoWork Corner

Claude CoWork, day to day

Quiet day in CoWork

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 Tier 1 candidates. The useful workflow to revisit is the Claude Desktop MCP bridge: Anthropic’s Desktop Extensions are described as one-click MCP server installation for Claude Desktop, which makes connector hygiene worth checking before a busy production week. Keep the active connectors small, named clearly, and tied to one purpose so CoWork does not become a noisy tool drawer.

T1Anthropic Engineering Blog07/23

GPT Desk

OpenAI, ChatGPT, Codex

Test it

Use Presence as the shape for a lead-triage agent

OpenAI introduced Presence for trusted enterprise voice and chat agents. Adrian should care because Goodsense, SET, and Street Attack Japan all have moments where a slow first response can lose value: inquiries, partner questions, location requests, and production follow-ups. The action today is to write one intake script for Goodsense: qualify the request, capture deadline and budget band, ask for reference material, then hand the summary to a human before any promise is made.

T1OpenAI News07/22

Use it

Turn ChatGPT Work into a one-hour client setup drill

OpenAI launched a ChatGPT for Small Businesses program to help entrepreneurs build AI skills, automate work, and grow with ChatGPT Work. Adrian should care because Grey Group can convert that theme into a practical offer for Japan-based small teams that need usable workflows, not AI theory. The action today is to build a one-page checklist: inbox triage, proposal drafting, meeting notes, customer FAQ, and weekly reporting, then run it internally once before selling it.

T1OpenAI News07/21

Small Money Systems

Small, repeatable, real

Test it

Sell a production inquiry revival agent

System: build a lightweight voice or chat follow-up agent for old production and brand inquiries. Customer: small agencies, event producers, and overseas teams trying to activate Japan leads. Offer: a cleaned lead list, scripted outreach, human-approved replies, and a weekly opportunity summary. Price: 75,000 JPY setup plus 30,000 JPY monthly maintenance. Existing assets: Grey Group OS, Goodsense production knowledge, prior proposal patterns, and Japan-English workflow. AI workflow: ChatGPT drafts scripts and summaries, with Codex or Claude Code wiring the intake form and status sheet. First action: pull ten stale inquiries and classify them today. Repeatability: the same pipeline runs every month on new dormant leads. Effort: half day. Expected value: recovered conversations and paid production leads without starting from cold outreach.

T1OpenAI News07/22 T1OpenAI News07/16

Test it

Package AI-assisted event recap clips for sponsors

System: create a repeatable AI-assisted recap package for events and activations. Customer: local sponsors, agencies, venues, and brand teams that need fast proof of activity. Offer: one short recap script, three visual directions, social captions, and an edit brief that a human editor can finish. Price: 55,000 JPY per event, with optional 25,000 JPY refreshes. Existing assets: Street Attack Japan footage patterns, Goodsense decks, production planning templates, and sponsor language. AI workflow: Luma-style creative agent references inspire shot and treatment variants, while ChatGPT writes captions and Claude Code assembles the delivery checklist. First action: choose one past event and make the package outline. Repeatability: each new activation reuses the same intake, edit brief, and caption structure. Effort: one hour. Expected value: small recurring production revenue and better sponsor follow-up.

T1Luma AI News07/23

Build Next

Deployable now

Use it

Build a coding-agent cost ledger

What to build: a simple repo-level ledger that records which coding lane handled each task, such as Copilot, direct API, Codex, or Claude Code, plus outcome and rough cost bucket. Why now: GitHub is making Copilot economics easier to compare against listed API rates. Effort: one hour. Expected impact: fewer expensive agent runs on mechanical tasks and cleaner decisions about which tool to use. Dependencies: existing GitHub repos, access to the coding tools already in use, and one shared task log.

T1GitHub Blog07/22

Test it

Create a video-model shot test board

What to build: a small test board with one brief, one reference image, one motion instruction, one generated output, and one pass-or-fail note per video model. Why now: Runway is positioning Gen-4.5 around motion quality, prompt adherence, and visual fidelity, which are production-relevant claims. Effort: one hour. Expected impact: better creative judgment before promising AI-generated production assets. Dependencies: access to the chosen video tools, three approved test prompts, and a folder for outputs.

T1Runway Research07/23

Try This Today

One action, right now

Pick three small coding or ops tasks today. Before running each one, choose the cheapest credible lane. Afterward, log tool, time, outcome, and whether a cheaper lane would have worked. This starts the cost ledger without building anything complex.