When everyone has the same AI, what do you differentiate on?

Tim Richardson opened Offcuts Live with a show of hands. Who's published something in the last year they'd now privately call slop? He put his own hand up first. Not many were brave enough to follow.

It was the second live recording of Offcuts, Tim's podcast and newsletter on where commerce meets AI. Forza Wine in Soho, a room full of brand and agency people, and us, as one of the sponsors.

Tim's setup was simple. Every brand in that room is using the same models now. When everyone has the same tools, the tools stop being the advantage, and he had a name for it: the flattening.

Then he handed it to three people with very different reasons to care.

Eleanor Warnock, a former Wall Street Journal and Sifted journalist who's now building her own media company. Natalie Varma, co-founder and CPO of Parfumerie, an AI-native fragrance discovery platform, after years at Farfetch, Net-a-Porter and Condé Nast. And Max Ottignon, co-founder of Ragged Edge, the London branding agency behind Wise, Monzo and Granola.

Their short answer: taste, voice and the decisions a brand makes for itself. Mine, after 55 minutes of listening: the same goes for the systems behind the operation, and brands have been fighting that flattening for a lot longer than AI's been around.

In this episode

0:00 Welcome and sponsors
1:56 The flattening: what's left to differentiate on?
3:37 How do you stop creating AI slop?
4:07 Will AI ever have taste?
7:46 Should brands let AI write for them?
14:33 A luxury brand's voice through an AI assistant
18:28 How do brands win when AI models converge?
27:04 Founder stories AI can't pull out of you
32:18 What happens to discovery when AI does the choosing?
37:14 Who are the curators now?
40:13 Good brand building is deciding what not to do
42:01 Audience Q&A

"Generative AI is an amalgamation. It's the average of the inputs. I don't think it's going to be able to achieve taste or copy taste."

Eleanor Warnock, journalist

Is taste the last thing AI can't copy?

Tim asked each of them for a number. One meant AI will match human taste soon enough. Ten meant taste is the only differentiator our species has left.

Eleanor went straight to 10. "For me, taste is all about something that is completely novel and totally different, and I actually don't think that a lot of people and a lot of organisations have real taste."

Natalie leaned the other way, with a more careful point than a number. AI already understands preference. Taste is different ("richer, more emotive, more context-based") and she thinks the AI that gets there will be built deep inside one category.

Max sided with Eleanor and kept it practical. "If you give a designer access to AI tools, they'll make a better design than somebody who's not a designer, and that's because they understand what good looks like."

Should brands let AI write for them?

Eleanor's own process is a decent answer. AI pulls the best quotes from her Granola transcripts and suggests headlines ("I'm so bad at writing headlines"). The writing itself stays hers, because her voice is what people come to long-form journalism for.

Asked what slop looks like when a client puts it in front of him, Max said: "I guess it kind of looks right. But kind of sh*t...". Ragged Edge uses AI tools all the time, with one test: is it better than what they'd make by hand? If it isn't, they don't use it.

Natalie spent years at style.com, turning Vogue's editorial voice into somewhere to shop, so she knows luxury voice was never just words. It's the visual codes, the cultural references, how the floor feels when you walk into the store. Boil that down to an AI assistant with the right adjectives and you've lost most of it.

How do brands win when AI models converge?

Max had written about this for Adweek- the models are converging, which leaves the AI companies competing on brand.

OpenAI, he said, has 600 creatives on staff. "So they must agree with me."

Granola is the one he knows best. Every decision on that brand started from one question: "What is the opposite of tech slop?" The answer was a hand-drawn G that's deliberately asymmetrical, a typeface based on co-founder Sam's handwriting and collage made by hand.

Eleanor came at it from the founder's side. She calls her approach Socrates as a service: sitting down with a founder and asking question after question until the real reason they started the business comes out. AI, she said, "does not do a good job of pulling those stories out."

What happens to discovery when AI does the choosing?

Natalie's whole business sits on this question, and fragrance makes it a hard one. "You can't smell the internet." Even so, she said close to half of fragrance is now bought online, a lot of it blind.

Her view is that general AI flattens a category like this. So Parfumerie maps what's underneath: notes, families, perfumers, even the bottles. Users build a fragrance wardrobe and can ask Iris, its fragrance assistant, anything.

Max had the last word before questions, and it's the one I kept thinking about on the way home. Tim asked how much of good brand building is deciding what not to do. "It's almost everything."

"You have to decide what you stand for, and you have to say it over and over again. And that means deciding what you don't say as well."

Product brands have been fighting this flattening for years

The morning was about the customer-facing side of AI: search, assistants, content. AI inside the business, in the back-office, is more our focus here at Nolo. The flattening being discussed however felt very familiar.

Product brands have been living with it since the first wave of SaaS tools for e-commerce, and the big legacy ERPs and PLMs before them. Those systems were built for the category average and sold to everyone. Eleanor's line about generative AI fits them pretty well: the average of the inputs.

Some of that is fine. Shopify, email, payments and shipping work much the same everywhere, so buy good tools and move on. How a brand comes up with product ideas, costs and buys them, and ultimately brings them to life is a different story, because no two brands do it the same way.

Squeeze that into a generic system and the brand slowly starts working the way the software does. Or the team gives up on it and goes back to spreadsheets. Nobody decides to become generic.

AI is just the latest round, and this time it's reaching the parts of the brand customers actually see. It's also made building your own custom systems cheap, which makes build vs buy a harder call. Max's point about taste applies here too: the advantage goes to whoever knows what's worth building.

His closing answer works just as well for operations. Decide what's most distinctive about how you make product, keep what's standard and build what's yours. Get the data underneath it right too, because clean foundations make AI a multiplier and fragmented ones make it a noise generator.

Your customers never see your operating system. They see everything it decides.

Big thanks to the ROIROI team for putting the morning on, and to Eleanor, Natalie and Max for a properly good conversation. The full recording is above, and it's worth the 55 minutes.