Seventy commerce and brand leaders, one Soho restaurant and a panel that couldn't agree on whether AI will ever have taste. Here's what we took from the morning, what it means for consumer brands, and where the conversation goes next: AI search, in January.

The company that built ChatGPT employs around 600 creatives. Max Ottignon put that figure to a room of brand and commerce leaders at 9am on a Thursday, and it cut straight through the idea that AI is replacing creative teams.
It came partway through our second Offcuts Live, held on 1 October at Forza Wine in Soho. Over breakfast, around 70 people from brands including Huel, House of Hackney, Varley, Derek Rose, Anya Hindmarch, Lucky Saint, PlayStation and Expedia sat in on a live podcast recording, then stayed for an hour of networking that ran well past its finish time. All 13 guests who left a review gave it five stars.
The full conversation is out now: Listen to the full episode. And if you want to be in the room for the next one, get first access here.
We called the session "The Great Flattening". Every brand now has access to broadly the same models, the same generative creative tools and the same personalisation engines. If the tools are the same for everyone, the tools stop being the advantage. So what's left to compete on?
Tim Richardson, our CEO and the founder of Offcuts, put that question to three people who come at it from very different places:
We chose them because we expected them to disagree. They did.
Eleanor opened the debate with the strongest claim of the morning. Generative AI produces the average of what it has learned from, she argued, which makes taste the last real differentiator a brand has. In her view taste is closer to cultural status than to a skill, and status is exactly the thing a model can't manufacture.
Natalie's reply was immediate: "I think quite the opposite actually." Her company is built on the bet that most AI has become very good at retrieval without yet solving for taste, and that taste is a problem worth solving.
The panel didn't resolve it, and we don't think anyone should pretend the question is resolved. What matters for a brand is that neither answer changes what to do this year. If Eleanor is right, taste is the moat and needs protecting. If Natalie is right, taste becomes something AI can learn, and a brand will need a very clear sense of its own taste before it can teach a system to apply it.
Either way, the work starts in the same place. A brand that can write down the decisions that make it recognisable, such as what goes into the edit, how the imagery feels and the things it would never say, is in a far better position than one relying on instinct that lives in two people's heads. Those decisions stay with a named person. AI can then speed up everything around them.
Max's description of AI slop isn't printable here. The point underneath the laugh was a serious one. Most people can't say exactly why a piece of AI-generated creative feels wrong. They just know it isn't good, and they trust the brand behind it a little less.
That gap between noticing and explaining is where brands get caught out. A team producing product descriptions, lifestyle imagery or email copy at volume will usually judge the output on whether it's accurate and on time. Customers judge it on whether it feels like the brand. Those are different tests, and only the second one shows up in conversion.
This is also where the 600 figure lands. If the business building the models still employs hundreds of people whose job is creative judgement, it says a lot about where value sits. The model makes competent work available to anyone. Distinctive work still takes people who know what good looks like, and Max's view is that this is changing what his whole industry is for.
For commerce teams, the practical step is to review AI output the way a customer would. Speed is only a saving if the result still sounds and looks like the brand.
One of the most popular ideas in brand marketing is that a brand should behave like a person: warm, witty, a little unpredictable. Max pushed back hard. People are inconsistent by nature, and that inconsistency is part of what makes them likeable. Brands earn trust by being relentlessly one thing.
His illustration stuck with the room: "A human answering every question with 'just do it' would be quite tedious."
This matters more now than it did two years ago, because AI is increasingly the part of a brand that talks to customers directly. Chat assistants, customer service agents, personalised emails and product recommendations all speak on a brand's behalf, often thousands of times a day. Giving each of them a quirky personality is tempting. Max's argument suggests the better investment is one clear voice, defined tightly and applied everywhere, so that a customer gets the same brand whether they're reading a homepage or asking a chatbot about a return.
Eleanor gave the clearest example of the morning of what AI can't yet do. Part of her work involves sitting with founders and investors and drawing out their stories. She calls it "Socrates as a service": asking question after question until the real reason someone is building their business comes out.
She described a founder who pitched her his app's AI personalisation features. Her response was to ask why he was the person building it. The answer turned out to be deeply personal, and it became the story that gave the brand its credibility. No amount of prompting would have surfaced it. It took time, trust and a conversation over food.
Eleanor isn't anti-AI. She uses it for research, transcripts and headlines. Her point is about sequence. AI can write a founder story well once it has the raw material. It is poor at finding that material in the first place.
We see the same pattern in our own work. Brands get the most out of AI when it's treated as a collaborator with a clear brief, and the brief comes from people who have done the human work first. Before handing brand copy to any tool, the most useful hour a team can spend is interviewing its founder, its buyers or its longest-standing customers. AI becomes far more useful once it has something true to work with.
The conversation ended somewhere nobody expected. Max argued that our obsession with efficiency has a cost. Friction is what makes us notice things, and when every moment is smoothed away, time seems to pass faster. Tim's sign-off: "Increase the friction, live longer."
It was a light note to finish on, but it applies directly to retail. A lot of AI investment in commerce is aimed at removing steps from the customer journey, and much of that is worthwhile. Some friction, though, is the experience itself. Unboxing a considered package, booking a fitting, waiting for a drop, browsing a shop that rewards a second look. A brand using AI to make its journey faster will need to decide which moments should stay slow on purpose.

Plenty of brands are building communities and events of their own, so here is what London taught us about putting the right people in a room.
"Great venue, interesting content and considered attendees for networking. 10/10." Simon Cockcroft, COO, CGC
"It changed the way I think about marketing and AI." Kate Afanasyeva, Google for Startups
"Top content and networking topped off with an excellent bacon muffin." Ross Newton Dunn, RND Digital Commerce
Watch the full episode here:
Next up: AI search, January
If AI produces the average, the next question is how a brand gets found at all when AI is doing the searching. Customers are starting to ask assistants what to buy rather than scrolling a page of results. That shifts how products are discovered, compared and chosen, and most brands' plans haven't caught up yet.
That's the subject of our next Offcuts Live in January. Same format: breakfast, a live recording and people worth meeting.
Thank you to our partners Centra, Nolo and Harper Concierge, and to Forza Wine.