What do OpenAI’s dots and Meta’s Muse mean for brands?
Flavia Barbat and Thomas Marzano in conversation. Legible and Lovable, episode 5. Published by Brandingmag, 2026-10-08. 62 minutes.
Episode notes by Thomas Marzano.
They make the personal AI agent real. OpenAI’s dots and Meta’s Muse work toward their owner’s goals around the clock, across devices, in apps and on the web. For brands, that changes where the brand gets used: summoned by a customer’s agent in the middle of a conversation, and rendered by customers onto themselves. A brand now has to be callable by agents, and its taste has to travel.
My day starts with my agent
My day doesn’t start with macOS any more. It starts with my agent, in Claude or in ChatGPT, and everything happens through that interface. Since February my agent has organised my files and folders. I no longer know where they are; I ask it to pull up the presentation, and it does. When I cut short clips from this podcast, I connect my agent to DaVinci Resolve, a professional post-production tool, and watch it edit the material down, find the quotes, build a storyline and make the transitions.
Our manifesto described this agent as a Jarvis: a butler in your pocket that knows you intimately, persists, and follows you from device to device. In September 2026 it shipped twice. Meta launched Muse on 8 September, aimed at consumers, in its own app and in WhatsApp. On 29 September, at its DevDay, OpenAI launched dots: always-on agents that learn their owner’s preferences, work toward goals on their own cloud computer, and pick up wherever their owner is, on a laptop, a phone or a voice call.
The story arc matters. Peter Steinberger released the open-source agent that became OpenClaw late in 2025 and joined OpenAI in February. I read dots as that idea made scalable and secure, although OpenAI has not drawn the line itself. Between the two launches, I see OpenAI as better placed: it is building anew, while Meta is fitting the technology into products it already had.
What these agents ask of us is a manager’s skill. A leader delivers through others: sets the goal, defines what good looks like, trusts the experts and steers at checkpoints. Now everyone gets that position, at the head of a team of agents.
The same day brought the second signal. OpenAI reworked the plugin directory in ChatGPT: developers publish plugins, and ChatGPT recommends them in the middle of a conversation. A plugin gives the agent new arms and legs. Connect one and, like Neo learning kung fu in The Matrix, the agent can suddenly edit video. To me this is becoming the App Store for agents, with the agent as the operating system. For a brand, that means being summoned in conversation. Expose what can be acted on, through a plugin and through WebMCP on your own site, and the customer’s agent can instantiate the brand in real time, in their name.
The third signal looks like a gimmick. For Fashion Week, Google Gemini worked with the stylist Sydney Bordonaro on Instagram tiles: a look, its garments and a short prompt. Screenshot the tile, upload it with a selfie to an AI image generator, and you are wearing the look. The stylist curates the taste, people try it on and share it, and the products travel the social graph rendered on people, well beyond the brand’s own audience, as Flavia pointed out. That is the Legible-Lovable Law closing its loop: an experience triggered by someone whose taste people trust, shared as a footprint that becomes legible as proof. What is being sold is taste.
Europe will wait. Muse is rolling out in the United States, and dots for Pro subscribers excludes the European Economic Area, Switzerland and the UK. The manifesto’s horizons are arriving on course, in their first versions.
The direction of travel is still clear. Regulation will decide the speed.
Questions this conversation answers
What can a personal AI agent like OpenAI’s dots actually do?
Work toward your goals without waiting to be asked. A personal agent such as OpenAI’s dots understands your goals through your context: it reads your calendar and your mail and proposes the next step, a draft reply, a meeting moved. It uses the tools you have connected, the plugins and connectors, and with your permission it reaches your apps and your computer. It lives in the cloud with its own computer, so it keeps working when you close the laptop, and you pick it up on your phone or by voice. In the launch video Flavia described, a dot proposes moving a finance meeting because its owner’s daughter has a recital: a chief of staff for your life. Teams will run several, a copywriter dot or a chief-of-staff dot, and OpenAI has specialist dots in preview for enterprises. Meta’s Muse does much the same for consumers.
How does a brand get discovered inside an AI conversation?
By being summonable. In ChatGPT, plugins are now recommended inside the chat: say you want to track your calories, for example, and the assistant can suggest a plugin that does it and offer to install it. So if you expose what can be acted on, your products, your service, your capability, you can be summoned and discovered in a conversation the customer was already having. That makes verbal branding matter in a new place: the name and description an agent matches against an intent. It also means accepting that agents act on your surfaces on behalf of their human, within permissions the user grants. For agents that browse rather than install, the same logic runs through WebMCP on your own website, which I cover in episode 4.
Listen from 42:29 · Episode 4: why AI recommends some brands · The Shortlist Effect
Is the AI agent becoming the new operating system?
I believe so, and you can already work that way. Once an agent is available on every device, operated by voice, and keeps working when you switch off the screen, it is where your day starts. Plugins become its app store. Flavia put the consequence in one line: people stopped going to websites, and now they will stop going to apps. My own day starts in Claude or ChatGPT, not in macOS, and I let everything happen through that interface. For brands, the surfaces they spent decades building, the site and the app, become places an agent visits on someone’s behalf. The manifesto described exactly this collapse of the distance between intent and fulfilment; with persistent agents and a plugin directory, it now has its machinery.
What devices will people use to talk to their AI agents?
Many, before a few win. The inputs and outputs will be visual, audible and tactile, and they can take almost any form: glasses, a ring, a speaker in the wall. OpenAI is working with Jony Ive, whose hardware company io joined OpenAI in 2025. Meta has announced Ray-Ban Meta Audio, its first audio glasses, shipping in October, and is bringing Muse to its AI glasses. I expect plenty of flops along the way. We played with MP3 players for years until the iPod defined the form factor. New modes will appear too: a household could run a family agent alongside each person’s own. For brands, the lesson is to stop designing for one screen and make the brand legible to whatever device the agent speaks through.
What AI skills should a brand team build, if not prompting?
Thinking skills. Two years ago you would have built a prompt library; today I tell clients to throw it away, because the models no longer need it. What teams need is cognitive capability. First, strategic thinking: can you articulate the goal, and say what good looks like and how you will measure it? Second, systems thinking: the sequence of steps, what each step needs, how the pieces fit, so an agent can be orchestrated toward the goal. Third, creative thinking, because the limit of what you can do with AI is now the limit of your own creativity and your ability to express it. And over all three, the judgment to evaluate what comes back and steer it. Forget prompting; build those.
Does deep expertise still matter when AI can do the craft?
Yes, and for some people it deepens faster. Flavia described her editorial work: by testing and re-judging many more variations, for many more clients, she gains experience she would once have needed years to collect. A writing tool that explains its edits is a door; walk through it and you get better. Domain expertise stays valuable. What AI exposes is a different line, between craft and creativity: there are brilliant studio musicians, and then there is Paul McCartney or Jimi Hendrix. Each brand has to decide which craft it keeps. Brandingmag keeps its editorial craft, because that is its product; for this podcast, the human authorship I protect is the live conversation between Flavia and me. Automate in support of the core you are valued for, and keep the core human.
What did Gemini’s Fashion Week try-on activation show about AI and brands?
That a marketing activation can now let people wear the product before they own it. For Fashion Week, Google Gemini and the stylist Sydney Bordonaro published Instagram tiles, each showing a look, its garments and a small prompt. You screenshot the tile, upload it with a selfie to Gemini or another image generator, which reads the prompt from the screenshot, and you see yourself in the look. People share the result, so the conversation spreads. Flavia saw the luxury case at once: a twelve-year-old who sees herself in a designer look may carry that relationship for years, until she can afford it. The same mechanic works for an interior designer’s material palette applied to a photo of your living room. It still relies on a screenshot, and it will get smoother.
What happens to a brand when customers render its products on themselves?
The products start travelling without the brand. They move through the social graph rendered on people, no longer as the asset on your website. That is the real-time rendered brand our manifesto described, in a first version. What is being sold is taste: a stylist or designer can codify taste in a look book or a style sheet, make it available as a prompt, and let people apply it to their own world. The brand is served through the curator’s filter. It also closes the loop of the Legible-Lovable Law. The meaning a brand codifies has to be verified by lived experience; here an experience is architected without a physical one, shared, and left behind as a footprint an AI can read as proof.
Calls made on air
Published 8 October 2026; windows added in October 2026. Each call as I made it, with the window in which I expect it to land: the year it could first happen, and the year by which I think it will have happened.
- The personal agent becomes your operating system. (25:19) From 2026, by 2028.
- Plugin directories for agents will compete with the App Store. (24:37) From 2026, by 2028.
- MCP and WebMCP will be on brand agendas within the next year. (45:53) From 2026, by 2027.
- AI devices go through many failed form factors before one defines the category. (28:23) From 2026, by 2028.
- Households will run a family agent alongside personal agents. (30:00) Direction of travel.
- Regulation sets the speed of personal agents in Europe. (59:24) Holds now.
In their words
“We’re all becoming managers in a way of a team of agents that you then need to learn how to orchestrate.”
“My day starts with my agent.”
“It’s 2026. Do you know where your files are?”
“I don’t feel like I’m losing my expertise.”
“What you’re selling is taste.”
Chapters
- 00:00 An “observations” episode thanks to three signals from the field
- 02:19 Jarvis, from sci-fi to OpenAI’s dots
- 05:00 What a personal agent actually does
- 07:40 Becoming managers of agents
- 13:09 Faster feedback, and the human who stays in the loop
- 18:12 The marketplace: an app store for agents
- 24:42 When the agent becomes the operating system
- 30:14 Forget prompting: these are the skills that matter now
- 33:46 Craft, creativity, and what not to automate
- 42:29 Discovery and WebMCP
- 47:26 Gemini at Fashion Week: taste you can try on
- 50:57 The real-time rendered brand, and what comes next
Sources
- OpenAI, “Introducing dots”, 29 September 2026
- OpenAI Help Center, “Getting started with your dot” (availability by market)
- OpenAI, “DevDay 2026 Recap”, 29 September 2026
- Meta, “Introducing Muse: The World’s First Personal AI Agent Built for Everyone”, 8 September 2026
- Meta, “Introducing Ray-Ban Meta Audio and More AI Glasses Styles”, 23 September 2026
- OpenAI, “A letter from Sam & Jony”, May 2025
- Wikipedia, Peter Steinberger (programmer)
- W3C Web Machine Learning Community Group, WebMCP, Draft Community Group Report
- Brandingmag, episode description on YouTube (Gemini’s Fashion Week activation with Sydney Bordonaro)
- Brandingmag, “Brand Constitutions: The Legible-Lovable Standard for Building Equity in an Agentic Economy”, November 2025
Where to start
When a customer’s agent can summon a brand mid-conversation, the question is what it finds: capabilities it can act on, and meaning it can infer. That is where the Agentic Brand Readiness Assessment begins.
Explore the engagements · Start a conversation
About the series
Legible and Lovable is Brandingmag’s podcast on agentic branding, hosted by Flavia Barbat, editor-in-chief of Brandingmag, and Thomas Marzano. In Brandingmag’s words: “Thomas identified the field’s dynamics and wrote the initial thesis; Flavia named the field and edited the content.” Episode 5 is an observations episode: three signals from the field, read against the manifesto’s horizons.
Where this sits. Agentic Branding is the discipline of making brands intelligible, governable, and desirable in an AI-mediated world. Its governing principle is Thomas Marzano's Legible-Lovable Law: brands must be legible to machines and lovable to people, with the machine now carrying both halves.
All episodes of Legible and Lovable. Episode notes by Thomas Marzano, updated 2026-10-06.