How fast are AI agents really changing the way customers choose brands?
Flavia Barbat and Thomas Marzano in conversation. Legible and Lovable, episode 1. Published by Brandingmag, 2026-09-16. 68 minutes.
Episode notes by Thomas Marzano.
Faster than we forecast, and slower than the headlines suggest. The manifesto put the personal AI agent in 2027; it shipped in the first half of 2026. Launch is not yet adoption, but brands now face an agent that is already available. The brands that hold up will be the ones whose value an agent can read, and whose conduct matches it.
“It’s here. It’s scary, but it’s here.”
Around the turn of the year, I installed the open-source personal AI agent now called OpenClaw, as people everywhere were buying Mac minis to run it. Suddenly I had an agent with a personality, a character I could speak to. My reaction was immediate: it’s here. It’s scary, but it’s here.
Flavia and I had placed that moment in 2027 when we wrote the manifesto. By the first half of 2026 it was shipping: agents that sit on the customer’s side, build a memory of who you are, act on your devices and go out onto the internet for you. A butler that knows you intimately and takes action on your behalf. Adoption is a slower story. The rollouts start on the newest devices, and I expect critical mass well into 2027.
What the agent changes is the distance between intent and fulfilment. You speak and it happens. I ask for a playlist for my run, and it is simply there. Every interface built to bridge that distance starts to dissolve, and with it the surfaces brands spent decades optimising. Whatever can be commoditised will be.
What survives is value an agent can infer and conduct it can verify. Beyond price and specification, an agent infers what a brand stands for, what choosing it says about the person choosing, how it looks and feels in their life. That meaning has to be codified, or the agent cannot find it. And it reads conduct: every gap between what a brand says and what it does leaves evidence, surfaced with every query. That makes this a governance question before it is a GEO question.
Most organisations cannot rebuild their governance in a quarter, so I advise working both ends. Make the brand cohesive now, everywhere it shows up, because inconsistency hurts more than a slightly-off strategy expressed consistently. And start building the governance for what is coming, in parallel.
The agent is here. Whether it can read your brand is still yours to decide.
Questions this conversation answers
Who will own the customer relationship when AI agents handle the checkout?
Whoever owns the distribution layer. The AI labs have extraordinary models, but they lack consumer distribution: Google and Apple own it, through Android and iOS. That is why I read the labs’ recent moves as a split. Anthropic, Microsoft and OpenAI are doubling down on enterprise knowledge work, while Google and Apple pursue the consumer companion that lives inside your phone. When OpenAI pulled back from its own checkout, I read it as strategy rather than a verdict on agentic commerce. For brands, the practical point is that the agent at the checkout will most likely arrive through the operating system people already carry, and mass adoption will follow the platforms with the broadest reach.
Do we still need an llms.txt file?
It does no harm, but it is not where the outcome is decided. Two years ago, when models were less capable, that kind of scaffolding helped them. Models now make sense of unstructured content well, and they will only get better at it. This site publishes one, because it is cheap and it helps today. What we do not yet know lies elsewhere: the governance layers platforms will put on top of their AIs, the standards for interoperability, and the regulation that will follow. Those are the variables brands control least. What brands do control is whether the content an agent reads is accurate, consistent and backed by proof.
Which brands can AI not turn into commodities?
The ones that live in rituals people choose. Everything that can be commoditised will be: when an agent can meet the specification and the price, it will. What it cannot replace is what people decide to spend their time on because it means something to them. The Bialetti moka on the stove in the morning. A vinyl record played from start to finish. Brands like these transcend the surfaces they appear on. The same holds in B2B, where the brands that host the culture and community of their field, and take an active role in bringing that community together, carry meaning no specification sheet can. Brand-building attention moves away from the surfaces we optimised for decades and into lived experience.
What does an AI agent need to understand about a brand beyond price and specs?
Its value, understood as an exchange rather than a list of values. An agent infers four kinds. Functional value: the problem solved, and at what cost. Aesthetic value: how it looks, feels and fits into a life. Existential value: what choosing it says about the person choosing. Normative value: the worldview the brand shares with its customers. When two brands solve the same functional problem, the agent breaks the tie on the other three, and it can only do that if they have been codified. That is the part I believe the industry has not yet reconciled with. In my work this is the Value Stack, and the Brand Constitution is where it is written down.
Why doesn’t GEO alone make a brand stand out in AI answers?
Because GEO works on the functional side of a brand, and that is the side every competitor can match. GEO and AEO were the first responses to arrive, and they made sense: the first question every company asks is how it shows up. Fixing the surface buys functional legibility. The agent can find you and verify the basics, but it learns nothing about why someone should choose you. My expectation is that how well a brand performs in an AI-mediated market will depend on how far it raises brand to governance-level discipline, and much less on how much GEO it has done or how many prompts it surfaces on. GEO belongs in the operational layer of agentic branding. It is not the heart of it.
What happens when AI exposes the gap between what a brand says and what it does?
The gap becomes evidence, and the agent surfaces it with every query and every interaction. I have long called it the integrity gap. Lately I describe the brands most at risk as costume brands: they wear a costume, and when it comes off, what is underneath does not match. Everything a company does leaves traces, and together those traces add up to its AI reputation. So the first job is to align the evidence models train on and crawl with what you want to be. That raises the hardest question of all: do you know who you are? Closing the gap belongs to the whole enterprise. Every CFO should be asking about it too, because the brand is one of the largest intangible assets a company has.
Should a brand’s AI transformation start top-down with strategy or bottom-up with cleanup?
Both, at the same time. I call it top and tail: take the safe bets that create impact now, and drive the bigger change in parallel. There is no silver bullet; the right mix depends on how a company is set up and what a team’s remit is. Category matters too. How exposed you are to AI mediation depends on technical complexity, price and how long people live with the product. A consumer-electronics purchase is a bigger decision than buying toothpaste. And if a team cannot touch strategy or governance, fix cohesion first, because that is the thing that bites the most. Inconsistency hurts more than being consistent to a strategy that is not yet fully sharp.
Calls made on air
Published 16 September 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.
- Consumer personal agents roll out on the newest devices first, with critical mass well into 2027. (16:34) From 2026, by 2027.
- Google and Apple own the consumer companion; the AI labs concentrate on enterprise knowledge work. (07:31) From 2027, by 2028.
- Brands will expose WebMCP tools on their websites, beside the pages made for humans. (18:05) From 2027, by 2028.
- AI becomes indifferent to how content is structured; scaffolding like llms.txt fades. (43:04) From 2026, by 2027.
- Whatever can be commoditised will be. (26:50) Direction of travel.
- Performance in AI-mediated markets will track brand-governance maturity, not GEO volume. (55:00) Direction of travel.
In their words
“You speak and it happens.”
“Launching doesn’t yet mean full adoption.”
“Everything that can be commoditized will be commoditized. It’s just as simple as that.”
“Yeah, because you need to know what that is in order to codify it.”
“Are you governing your brand as a piece of surface, or are you governing the brand as something institutional across your organization?”
Chapters
- 00:00 Looking back at the manifesto
- 05:11 Trust, checkout, and who owns distribution
- 10:26 When the predictions came true
- 17:28 WebMCP and the two internets
- 18:24 Do you still need llms.txt?
- 21:28 Why two perspectives make one discipline
- 24:19 Ritual in an agentic world
- 29:13 Earning the right to be named
- 34:40 When the intangibles become tangible
- 36:23 The four value pillars
- 41:05 Back to GEO and AEO
- 46:21 Governance and the integrity gap
- 56:42 Top-down strategy meets bottom-up cleanup
- 1:05:10 Closing thoughts
Sources
- OpenClaw, open-source repository
- IT Brief UK, report on OpenAI scaling back checkout in ChatGPT, 16 March 2026
- Google, “I/O 2026: Welcome to the agentic Gemini era”, 19 May 2026
- Apple Newsroom, Siri AI announcement, 8 June 2026
- W3C Web Machine Learning Community Group, WebMCP, Draft Community Group Report
- The /llms.txt proposal
- Brandingmag, “Brand Constitutions: The Legible-Lovable Standard for Building Equity in an Agentic Economy”, November 2025
Where to start
A readiness assessment starts with how agents read your brand today, from the functional basics to the value they infer, and with where your claims and your conduct part ways. 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 1 opens the series nearly a year after the manifesto, testing what held true and what moved faster than planned.
Where this sits. Agentic Branding is the discipline of making brands intelligible, governable, and desirable in an AI-mediated world. Its governing principle is the 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.