Where does human creativity go when AI agents carry the brand?
Flavia Barbat and Thomas Marzano in conversation. Legible and Lovable, episode 2. Published by Brandingmag, 2026-09-18. 74 minutes.
Episode notes by Thomas Marzano.
Upstream. When AI agents act in a brand’s name, in every interaction and at a scale no team can supervise, the brand can no longer be improvised downstream. The craft moves into the studio, where people decide what the brand is and codify it. On tour, the agents play that script faithfully, and a person stays accountable for whatever the script did not foresee.
The studio and the tour
By the spring of 2025 I had a clear picture of how AI would reshape the market. In April I called Flavia with one message: we need to talk, because I have something. We met in Rome in August, a couple of scorching days at 35 degrees, with hours over dinner and drinks. Flavia, hear me out. Over those days I described the field to her for the first time.
An agent-mediated shift was under way, in which brands would have to be legible to machines, governed, and loved by people. Guidelines that tell people to go approximately over there would have to become constitutions, governing what AI agents may do inside organisations that are themselves becoming agentic. And our craft, our creativity, would move upstream, into the curation of everything those agents learn from. Flavia had real reservations about what AI meant for the creative craft, and she said on air that the upstream idea was the key she had been waiting for. Between August and October we turned it into the manifesto, published in November 2025, and somewhere in that work she asked the question that named the field: aren’t we talking about agentic branding here?
The urgency comes from what I call technology coming of age. For thirty years, through home computing, e-commerce, social media and mobile, it was still us operating the machine: screens, keyboards, buttons, forms. AI puts an intermediary between intention and fulfilment. Brands stop being acted upon, by customers and by the people who manage them, and start acting themselves, in every interaction with every customer.
A guideline was designed to be interpreted by people. Agents need something firmer, a law they adhere to: a constitution that spells out behaviour, learns from the agents’ own actions and is measured against the brand’s acceptance criteria. When an agent meets a case it cannot handle with confidence, it escalates to a person, the case is documented, and the system improves. Accountability stays human, because there is no legal entity called an AI.
Flavia asked the question that drew out the rest: if guidelines were interpretable, where does the experimentation go? I answered with the analogy I have used throughout my career. In the studio, a band diverges. It explores, plays, writes through the night, then converges on an album. On tour, it executes a choreographed script, because the audience needs to get what it came for. Agents are the tour.
The studio is where the people who govern and curate the brand do the creative work, and once a decision is made, it is codified so the agents can replicate it and extend it. I always struggled with downstream flexibility; it produced inconsistency nobody asked for. Agents acting in the brand’s name in every interaction would multiply that inconsistency, and leave a record of it. And AI is still far better at replicating creativity than originating it, because it cannot have a human experience.
The same move applies to meaning. The emotional arc of a brand used to live in campaigns built for people, around personas and demographics. An agent can now infer what its user prefers, and it looks for a brand’s meaning where it can read it. So that meaning has to be packaged for inference as well as for people. For a brand leader, that changes where the effort goes: less time policing executions downstream, more time deciding, and writing down, what every agent will carry out.
The craft is moving into the studio. And one day we will drop the word agentic, the way we dropped digital.
Questions this conversation answers
Why are AI agents a bigger shift for brands than the web or mobile were?
Because in every earlier wave, people still operated the machine. Home computing, e-commerce, social media and mobile each brought extraordinary capability, but we still spent our time behind screens and keyboards, clicking buttons on operating systems, apps and websites. AI agents take over that operation. They grasp the intention and carry it out, which collapses the steps between wanting something and getting it. For brands, the direction of the relationship changes: brands were acted upon, by customers and by the teams that managed them, and now they will act, in every interaction with every customer, through agents. I no longer treat this as a signal. In May and June 2026, Google and Apple both announced assistants built into their platforms that know your personal context and take action for you: Gemini Spark and Android Halo from Google, Siri AI from Apple. The caveat is adoption. Both start narrow: Google’s with a limited group of users, Apple’s on recent devices only and not in the EU at first.
Do brand guidelines still work when AI agents act for the brand?
They were built for a different reader. We wrote a strategy, built guidelines on it, trained staff and agencies, and ran a brand desk to keep execution on course. Guidelines were always meant to be interpreted by people; the word itself implies approximation. Now a brand gains agency in two places. Inside the company, agents produce the presentations, communications and content that employees used to produce. Outside, the brand is exposed through an API or an MCP server that acts and delivers value in real time. Both have to be governed the way machines can be governed: with a constitution, the foundational law every agent adheres to. Flavia summed up what goes in it: we do this, and we don’t do this, with far more examples than any guideline held. And someone has to decide what goes in and what comes out as the brand evolves.
How do AI agents decide which brand information to trust?
By the standard people already hold their own agent to. The contract between a person and their personal AI is one of trust: if it keeps returning things that aren’t true, it loses them. So the agent’s incentive is to pass on only what it can stand behind. It needs fast access to a brand’s core truth, its products, services, pricing, provenance and availability, because it will not spend hours working it out. It checks whether the claims are accurate. And it checks whether the market echoes them: what a brand says about itself and what others say can differ, and the agent surfaces those nuances and caveats. Flavia put it neatly: its algorithm prioritises credibility. Making a brand intelligible, in the sense of our definition, means doing that work before the agent arrives.
Who is accountable when an AI agent acts on a brand’s behalf?
A person. There is, as yet, no legal entity called an AI, so a human will always be accountable for an agent’s performance and output. That accountability has to be baked into the systems we design. The agents that face customers should not experiment. When one meets a case nobody predicted, or cannot respond with full confidence, it escalates internally, and people catch it and steer it. The case is documented, it becomes part of what the system learns from, and the system improves. Governance here means a learning loop with human oversight, evaluated against the brand’s own acceptance criteria.
Can AI tell why a customer prefers one brand over another?
Increasingly, yes, and that is what is new. We already live with narrow AIs that track taste. Instagram, TikTok and Pinterest learn what captures you, which is why my feed looks different from yours. A personal agent brings that ability together with general intelligence and memory. Beyond whether a product meets the specification, it can infer the softer drivers: your aesthetic taste, what a choice says about you, how you see the world. Often the preference is already in the prompt. Few people ask an AI where to go on holiday with no idea at all; they say Italy, maybe Tuscany. They say look at Nike, but stay open to others. Brands that already hold a place in someone’s mind get written into the request. Flavia called the agent a matchmaker of values. It can only match what it can read.
Why doesn’t the emotional story in our campaigns reach AI agents?
Because it was packaged for people, and only in some places. Brands built campaigns around personas and demographics, found a few buckets of needs, and crafted the storytelling and the emotional arc for human consumption. Go to the same brand’s product page and those values have mostly gone: a key visual, maybe a headline. Two things now change. We can personalise to the individual, so it makes more sense to architect a brand’s meaning so it can be matched person by person than to keep working in big persona buckets. And we can no longer package that meaning for humans only. It has to be packaged so that an agent can infer it reliably and conclude: this brand will probably mean more to my user than that one.
What will brand experiences look like when interfaces are generated on the spot?
Brief, personal and generated in the moment of need. You can see it already. I connect Spotify to my AI assistant and ask for a 1980s playlist because I’m going running in twenty minutes. The playlist renders in the conversation, branded, goes straight to my Spotify, and the interface disappears; I never scroll through the app. Google has published an open protocol for generative interfaces, A2UI, and images already appear seconds after a prompt. Where this ends is everything generated in the moment of need. Flavia’s analogy is the right one: in a game, the world does not exist in full while you play; only your view is rendered, from rules set in advance. Brands become worlds like that, with characteristics, rules, an aesthetic and a way of behaving, and each user gets a version of the world that may never happen again. Our job is to define the world well enough that whatever renders is still the brand. Listen from 1:04:12 · Related: After the Interface
Calls made on air
Published 18 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.
- Personal agents will transact with brands without the customer in the room. (19:15) From 2026, by 2027.
- Mass adoption arrives through the operating systems people already carry. (23:26) From 2026, by 2027.
- Brands will be exposed as APIs and MCP servers that act on their behalf in real time. (32:57) From 2026, by 2028.
- Personal agents will infer the softer drivers of preference: taste, worldview, what a choice says about you. (48:05) From 2026, by 2027.
- Generative interfaces are imminent, and experiences will end up generated in the moment of need. (1:07:54) Generative interfaces from 2026, by 2027; the end state is the direction of travel.
- A human stays accountable for every agent’s output while no AI legal entity exists. (35:14) Holds while the law stands.
- Robots will instantiate brands, which the manifesto missed. (1:09:08) Direction of travel.
- We will drop the word agentic, the way we dropped digital. (1:12:51) Direction of travel.
In their words
“In all those different shifts, it was still us as human beings having to operate the machine.”
“A brand can no longer survive with a brand guideline that is sort of telling human beings, approximately over there.”
“Accountability is something that we cannot delegate to a machine.”
“Once you go off on tour, that’s a militaristic script that you execute.”
“That costume will break. You can no longer package it in that way. You have to be it.”
Chapters
- 00:00 Welcome and introductions
- 00:59 Technology “coming of age”
- 05:26 Rome, and the case for constitutions
- 10:14 Why human creativity moves upstream
- 15:04 The definition: intelligible, governable, and desirable
- 17:08 The proof it’s happening, and why adoption still lags
- 24:50 Intelligible: trust, accuracy, and the corporate knowledge graph
- 30:07 Governable: from guidelines to brand constitutions
- 37:24 Where creativity now lives
- 43:37 Desirable: two ways a brand wins
- 55:33 The integrity gap and what agentic branding is not
- 1:04:06 Two simultaneous realities of brands interacting with AI
- 1:08:44 Generative UI, robotics, and brands as worlds
- 1:13:00 Wrap-up and what’s next
Sources
- Google, “I/O 2026: Welcome to the agentic Gemini era”, 19 May 2026
- Apple Newsroom, Siri AI announcement, 8 June 2026
- Google Developers Blog, “Introducing A2UI: An open project for agent-driven interfaces”, 15 December 2025
- Spotify, “Introducing Spotify in ChatGPT”, 6 October 2025
- Brandingmag, “Brand Constitutions: The Legible-Lovable Standard for Building Equity in an Agentic Economy”, November 2025
Where to start
Before a brand can write its constitution, it needs to know where it is still being improvised downstream: the places an agent would have to guess. 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 2 is the series’ first foundational episode: where the term came from, what each part of the definition asks of a brand, and where the creative craft goes next.
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.