Earlier this year, I visited Casa Azul, Frida Kahlo’s home in Mexico City. I love seeing the homes of great creators and am always struck by how they lived seemingly ordinary lives, while crafting legacies of extraordinary beauty.

Amongst the art, the ceramics and the Pre-Columbian sculptures in the garden, was a digital display where you could type in a question and an AI version of Kahlo would answer. Here was one of the most mythologised artists of the twentieth century, apparently available for a chat — but hardly anyone was engaging with the device.

Take a famous person, ingest enough material associated with them and put a chat window on top. From Ask Ralph (Lauren) for styling advice to Ask (HBS Professor) Jeff Bussgang for feedback on your pitch deck, the basic proposition has been similar, but it may be changing.

The Ask Frida display at Casa Azul

From Persona to Proxy

A few days ago, on David Senra’s podcast, Sam Altman described the AI he actually wants.

Altman said that he increasingly feels constrained not by model capability, but by how little useful context they have about him. He wants an AI that can read the internal Slack posts he will never get to, absorb customer stories about where ChatGPT worked and failed, read research papers he hasn’t, and bring all of that information to bear when he has to make a decision.

His proposed agent needs to know things Altman himself may never have encountered at all. The shift is from an ‘archive of me’ to a ‘context system for me.’ From “Make me machine readable” to “Figure out what is useful for me in a machine readable world.”

The shift is from an ‘archive of me’ to a ‘context system for me.’ From “Make me machine readable” to “Figure out what is useful for me in a machine readable world.”

But a proxy that merely gathers more information is not enough. To be useful, it has to decide what matters to me — and that requires some model of my judgment. Companies like Continuum Twins and Delphi are trying to address this with “cognitive twins” that combine retrieval — what you know, have said and written — with a structured model of how you reason.

There is some research behind this. In a 2024 Stanford study, agents built from just two-hour interviews with 1,052 people reproduced their answers to a major social survey 85 per cent as accurately as the individuals reproduced their own answers two weeks later. The study did not prove that AI can replicate human judgment, but it did show that richer information about a person’s life and beliefs made the resulting model much better at predicting them.

But this means we must confront another, less discussed dimension of AI inequality.

AI cannot arm me with relevant knowledge, unless it can understand me. And for it to understand me, it needs data – a continuing stream of evidence about how I think and decide. A mediocre executive with 20 years of emails, presentations, Slack discussions and recorded meetings may be easier to augment with AI than a brilliant nurse whose judgment has been expressed almost entirely through practice.

AI cannot arm me with relevant knowledge, unless it can understand me. And for it to understand me, it needs data.

What does this new emerging inequality mean for people and organisations whose experience and judgment has been not been rendered machine legible?

Does it mean we should all start wearing lapel mics (or devices on our temples) and narrating our own competence into a transcript? I shudder to think of where this could lead. But there is no denying the fact that if the personal AI that Altman wants becomes a critical advantage, being well documented becomes a form of privilege in its own right — independent of how good the judgment being documented actually is.

Being well documented could become a form of privilege in its own right — independent of how good the judgment being documented actually is.

Person vs Proxy

Back to Frida. I thought hard about why that digital display felt so pointless, and realised it wasn’t because the AI knew too little about her. There’s probably enough material on her life to build a genuinely convincing digital persona.

The problem was that Casa Azul had already given me a rich, physical, contradictory encounter with her — and the chatbot collapsed all of it into a single banal proposition: Ask Frida a question. Even the most maximally documented life in the world, run through a chat window, produced something duller than the house that holds her bed and her monogrammed plates and her paint-stained clothes.

That’s the warning in this, I think. As we stand today, total documentation will not guarantee a faithful proxy of a human — it will only guarantee a proxy and the two are not the same thing.

Such a perfect form. Frida and Diego’s initials intertwined on the monogrammed crockery

This is an unexpected and fascinating AI story and it is also the rabbit hole I am currently inhabiting.

TLDR

Brunello Cucinelli, a storied Italian cashmere house, has built an AI company called Solomei Ai, which in turn has launched an e-commerce platform called Callimacus. Cucinelli’s new shopping site, built on Callimacus has no pages and no prescribed pathways. An orchestration layer of agents builds the experience as you go. The idea is to recreate the shop assistant, not just the catalogue. It left me wondering what it takes to encode brand and culture, so the system feels like “Cucinelli.” Has to be much beyond brand guidelines as we know them.

Here’s a deeper dive

Brunello Cucinelli, the Italian cashmere house founded in 1978, is based in an ancient village in Umbria. Following a 2019 visit from LinkedIn founder Reid Hoffman, who spoke about the changes AI would bring, the company began building a separate AI entity, Solomei AI. Five years later, Solomei launched Callimacus, an e-commerce platform named after the philologist who catalogued the Library of Alexandria.

In January 2026, Cucinelli rebuilt its entire e-commerce experience on Callimacus, throwing out many of the must-dos of e-commerce design and architecture. Their creative partner was design firm, Makemepulse.

Last week, Salesforce signed an agreement to invest in Solomei AI.

What does Callimacus do that is so different? And could this be a preview into the future of online experience?

Health Warning: The e-commerce site, shop.brunellocucinelli.ai is not available in all geographies. Readers in India will have to make do with the screenshots in this article. But I promise that this piece discusses questions beyond one brand’s site.

Founded on intent, not just history

Francesco Bottigliero, Chief of Humanistic Technology at Brunello Cucinelli and CEO of Solomei AI, says something we can all identify with.

“I buy batteries for my AirTags online. They arrive the next day. And then the merchant spends the following weeks sending notifications asking whether I’d like some more. I don’t eat batteries like M&Ms! Now suppose I go to a store to buy a cashmere sweater for my wife. When I return to the store, nobody asks me to buy more of the same cashmere sweater. They already know.”

Yes, we can build better recommendation engines but his point – that purchase history is often a poor proxy for present intent – is valid.

Callimacus infers probable intent from behavioural signals, instead of history alone, trying to ascertain if the user’s intent is product-driven, discovery-led, or inspirational, and adapts the flow and components accordingly.

Are you moving slowly through the films about how the cashmere is made? That’s a person looking for an idea. Twenty minutes later you’re clicking fast through product. That’s a person close to buying. The site shifts and composes content accordingly — more editorial for the first, more product for the second. The user can enter a prompt or just be prompted by the site, as it observes her/his behaviour.

Callimacus infers probable intent from behavioural signals, instead of history alone and adapts the flow and components accordingly.

Goodbye to the page as we know it

With categories, menus, filters, search boxes and grids, today’s e-commerce experience puts the burden of discovery on the user.

In fact, Bottigliero points out that the web page wasn’t designed for shopping at all. It came from print, and from CERN, Geneva, thirty years ago, when scientists were trying to find a way to share research papers between offices.

Callimacus has no fixed templates or predetermined navigation paths. It assembles what you see, as you go, in real time. The experience is quite different from the typical conversational commerce, where a bot is tacked on to a site.

Callimacus has no fixed templates or predetermined navigation paths. It assembles what you see, as you go, in real time.

No generic personalisation

Most personalisation you’ve experienced is a set of pre-written scripts.

Callimacus has no fixed journeys. A group of AI agents works out what you seem to be there for and builds the experience live, rebuilding it as you change your mind. Two people asking for the same blue sweater can end up with visibly different content.

The orchestration underneath is not trivial. The AI isn’t sitting in a box waiting to be asked something. When a system is composing the page in front of you, a two-second pause equals a broken website experience. Callimacus uses a combination of models, including open-source ones, on specialised hardware to ensure performance and speed.

Callimacus has no fixed journeys. Two people asking for the same blue sweater can end up with visibly different content.

Encoding brand and culture

Instead of merely digitising the catalogue, Callimacus is trying to digitise the shop assistant.

A good shop assistant is a carrier of accumulated judgment (I know, I know — I tried hard to find another word). This judgment comes from proximity to customers, to the team and to the brand. But machines don’t absorb organisational culture simply by being embedded inside an organisation.

Image Credit: Makemepulse

Building Callimacus almost certainly required making much of Brunello Cucinelli's tacit knowledge explicit. This generally gets swept under “training the model,” but as creative practitioners, it is worth dwelling on.

What we are talking about goes way beyond a brand book or a colour palette. This is organisational knowledge about what gets shown and to whom, what gets held back, and why.

For Callimacus to feel recognisably Brunello Cucinelli, decades of brand knowledge had to be translated into principles, examples and constraints the system could act on. We don't know exactly how Solomei AI did that. (We do know that their team included a philosopher and an artist)

For Callimacus to work, somehow 46 years of culture and brand knowledge had to be encoded in a form a machine could act on.

The obvious objection

I fed this article into Claude and ChatGPT and asked them to challenge it.

Both made the same counter-point that as models improve, no guidance or codification by humans will be required. I struggle with that. Even if they are right, the immediate need is for organisations to make their accumulated judgment legible to software.

Bottigliero says they have received 40 inbound queries for the Callimacus platform, most from outside luxury. It is far too early to say whether this will be a commercial success, or whether Solomei’s approach will survive at all. But that is besides the point.

For years, we've said that a brand lives in people's heads. Now brands will have to live in a form that machines can understand too.

For years, we’ve said that a brand lives in people’s heads. Now brands will have to live in a form that machines can understand too.

Especially where the difference between brands is qualitative, rather than operational, the ones that benefit most from AI will be those that can codify what makes them distinct — in a form an AI system can act on. (At least for now, as the LLMs remind me.)

Encoding brand and culture to make it machine readable is the rabbit hole I’m currently buried deep in. If you’re working in this space, get in touch.

“Let’s call it Cockroach,” said A.

“Don’t be ridiculous.” D shook his head, rubbing his eyes behind his spectacles. He had come with a deck – naming protocols, competitive landscape, desired tonality – the list was impressive.

“Just think for a minute,” said A. “We can turn it into something they never imagined.”

“It’s an insult,” D interrupted, his voice rising. “It means vermin. The thing you step on in the kitchen at night. You are forgetting the basic rules of naming – we are not naming a thing, we are naming a feeling, and the feeling here is disgust.”

“Guys.” N capped her quaint ink pen. “Let’s not squabble. It’s 2040 for god’s sake. All we need to do is ask BR-AI-N.”

At the sound of its name, the ovoid console in the corner glowed green. “Hello” it said “I’m here to help you.” Neither the console, nor the green light or the greeting for that matter, were necessary for BR-AI-N to work. They were artefacts from another era, but retained because they felt familiar.

BR-AI-N itself had been christened by a consultancy that had delivered a 2000 word rationale, of which the operative sentence was that the hyphens signalled human-machine partnership at the level of the glyph. BR-AI-N had been an unequivocal success. It had named six in every ten launches in the country that decade. And there had been a lot of new launches. New companies, new products, new drugs, new coalitions. The system required an expensive subscription, but what was the option? Every creative team had to shell out or go out of business.

“What mode would you like me to work in while recommending a name? Shall I be mainstream or contrarian?”

N had requested Shah Rukh Khan’s voice in BR-AI-N’s settings (it was also part of the free options) and she still smiled every time it asked a question.

“Both,” said N.

It took just a few seconds. Mainstream first. Yuva. Shakti. Sankalp. Uday. Each with a short, live survey and analysis attached. BR-AI-N had run each name past 40,000 synthetic citizens, aged them five years, then put them through two global wars, a bad monsoon and three domestic elections. The names on its list had endured. High recall, great phonetics, easy decoding, warm on second exposure, clean across 20+ language markets and resonance with the target demographic. It noted that Shakti had been deployed 1,100 times since Independence and that this was an asset rather than a liability. Familiarity, it said, is a form of trust. Suitable urls can be booked in the next 5 seconds. Logos generated and tested in the next 10 seconds. And as a bonus, BR-AI-N had already created a sonic identity for each option.

Next, BR-AI-N briskly moved to being contrarian. Kr-ai-anti, it said – signifying new age rebellion. (Note the clever pun, which will also make it easy to register.)

More contrarian options – Changemakers. Enough. All of these were unexpected names that would be refreshing to an audience fatigued by platitudes. These names signalled subversion, said BR-AI-N, but with a humane side.

The room was silent.

“Should I rank these in descending order of projected impact?” purred SRK / BR-AI-N, after waiting a few seconds for a response.

A leaned toward the console. “What about Cockroach?”

The light pulsed. “Is this a new task?” asked BR-AI-N in a friendly tone. Are you looking for the best pest control solutions?”

It waited for a response and then backtracked and produced the relevant analysis. Sentiment was strongly negative across all tested markets. Primary associations were infestation, disease, poverty, contamination. Nothing to redeem it visually. Just the sight of the bug made many people feel uneasy. BR-AI-N did not recommend proceeding. It flagged a category-entry-point risk.

Then, crisply, it explored adjacent areas and offered alternatives. Carapace (signals protection) Blatta, from Blattodea — clean, scientific, could be imbued with any brand vision.

“For god’s sake” muttered D under his breath.

“I can generate more options” said BR-AI-N helpfully. “Would you like me to do that?”

D and N looked at A.

“Nah” said A stretching, as he got up from his chair. “We’re calling it Cockroach.”

Travelling through four different cities last month, I spoke to design and communication teams at several companies and agencies, all complaining about the same problem.

The problem was called “But now you have AI”

Here are some of the stories I heard and I think you’ll relate to most of them.

The “Sasta Tikaoo” Website

Say website redesign and most marketers and designers will run in the opposite direction. It is that most painful of interfaces to change – a multi-headed hydra made more egregious by the fact that it is so critical. How then to explain to the boss in love with Lovable, that a 200 page content-heavy website cannot be developed “very quickly and with half of last-time’s budget.”

Yes, he created a site for his daughter’s 10th birthday in an hour last Sunday. But something with a multi-user CMS, SEO/AEO that survives a Google update and analytics that actually match the sales pipeline, doesn’t quite compare.

The Case Study Farm

Case studies are the lifeblood of B2B communication strategy. And everyone who has struggled to publish one knows how difficult it is to land a case study that actually influences the reader. Get alignment on the story between account and client teams. Don’t use the client’s name. Don’t use real metrics. Get forty approvals from both sides… the list is endless. Yet, now bosses are asking why a hundred case studies can’t be produced by Friday. Sure they can, as long as legal gets back before Q3.

The In-house Rebrand

This one came to me from a wounded agency head fresh out of a rebranding discussion with an infrastructure company.

Apparently the CEO said to the CMO: “After all, I’m not asking you to use Canva. But what’s your excuse now that there is Claude Design? Why do you need a specialist agency? Why can’t you just do this in-house?”

The CEO then sat back, visibly pleased. He was plugged in. He was ahead of the curve. Questions about the art and and science of brand positioning and visual identity were for people still stuck on Dali.

The Authentic LinkedIn Post

It is time for the “D” discussion. Sooner or later, every comms professional will be called into discuss personal branding for the C-Suite.

From Amodei to Altman, modern entrepreneurs control their own distribution, you will be told. And so you diligently start with a multi pronged analysis of the industry landscape. You think about your CEOs’ reputation and her personal style. You sweat to create a set of LinkedIn posts that are sharp, witty and interesting.

The posts come back to you 20 minutes after you submit them. The boss has put them through ChatGPT and now they sound like everyone else. You know you will not win this battle. All you can do is wait resignedly for someone to put them through Pangram.

Thinking back on my conversations, It is clear to me that there are two sides to AI adoption: The first is failing to adopt fast enough and the second is adopting AI mindlessly with no real value.

In both cases creative professionals are complicit.

For too long we have focused - and drawn our audience’s focus - to the glossy, finished output. There has been little talk of process, of exposing the innards of what it takes to create something meaningful and useful. This has come to bite us because AI can now produce glossy, finished output much faster than we ever could, even if it is ultimately worthless.

If I had one piece of advice to the founders, marketers, designers and writers reading this, it would be:

Map your workflows. Understand where AI can genuinely transform them – and there are many places it can. Segment these into low-hanging fruit, medium-term shifts, and the slow, but necessary thinking tasks that only a human can do. Prioritise ruthlessly.

Then sit down with your boss / client and walk her through the map. This sounds much easier than it is. It is a long, hard road to get an audience fed on a rich diet of “options” to think differently. But if you want to thrive and grow in in AI world, it is the only one worth taking.

Apologies to all bosses and clients who were hurt in the reading of this post

Last week an old friend called.

“Is something wrong,?” she began tersely.

Then, in response to my bewildered denial, and somewhat sheepishly, she said:

“You are the only one who didn’t like or comment on the photo I posted in the group. I wondered why.”

My heart sank. I knew exactly which photo she was talking about. The one of her daughter’s graduation. I’d seen it, thought “I’ll respond properly later,” and then completely forgotten about it.

I had, in the language of WhatsApp, said nothing. And in the language of WhatsApp, saying nothing is a position.

After taking a course in improv, I’ve been going down the rabbit hole of non-verbal communication - how to read a room, how to signal status, how to influence it. (More on that later.) What struck me, somewhere in the middle of all this, is that nobody talks about the version of non-verbal communication we practise every single day: the one happening in our phones, in groups we can’t leave and chats we can’t un-see. It is rich, it is loaded, and it has rules. Nobody wrote them down but everyone knows them.

Nobody has written down the rules for digital conversation. But everyone knows them.

The Shape Shifter

Some emojis signal pretty much the same thing, whether they are used on an open platform like LinkedIn / Instagram or a closed one like WhatsApp. Take the single ‘👍’ for example. Ubiquitous, understated and careful not to over commit, it says the same thing everywhere, “I see you, but must now move on to more important things.”

Other emojis, however, shift meaning as they travel. The single ❤️ on WhatsApp is typically interpreted as an enthusiastic like: To appreciate, but also to signal that a conversation has come to its natural end. A heart on a LinkedIn post is a whole other story. On LinkedIn, replying with a heart means you didn’t just find this useful. You found it moving. It says you feel things deeply and you are not scared to let the author of the post and your network know this. Add a 🥳, and now you’re practically gushing.

The Private Admirer

There is a specific type of person in every network who will slide into your DMs to say your post was brilliant, that it made them see things in an all new light.

But while composing their heartfelt private message, they have not liked the post. Not commented. Not reacted with so much as a 👏. They have consumed your content, appreciated it privately, and left your reach exactly where they found it.

You will respond graciously. You will say “Thank you, so glad it resonated.” What you will not say is: “Your admiration does not show up in my analytics.”

The Power Play

You sent a message at 6 PM.

The ticks went blue at 8.24 PM.

It is now the next morning. Still no reply.

The message has been received. The response is being rationed. The reply will probably arrive, breezily, at 11 AM the next day, as if the 17+ hour time lag never happened. Whether this is a flex, or a genuine lapse, you will never fully know.

The Heavy Lifter

The folded hands 🙏🏼 emoji deserves its own section because nobody can agree on what it means and everyone uses it constantly.

Is it gratitude? Prayer? A namaste? It can mean Thank you / Please / I’m begging you / I’m so sorry for your loss. This emoji does a lot of heavy lifting while refusing to be pinned down.

In fact, 🙏🏼 is less of a response and more a gesture in the direction of a response. It says “I acknowledge the weight of this moment” and that is more important than a narrowly defined meaning.

And maybe 🙏🏼 is so widely used because there is an increasing number of moments in our lives where acknowledgement is all we can honestly offer.

Maybe 🙏🏼 is so widely used because there is an increasing number of moments in our lives where acknowledgement is all we can honestly offer.

The Silent Exit

There is no action more theatrical than the quiet exit from a group. No message. No goodbye. No explanation. Just:

<This person> has left the group

It is a single line that everyone in the group reads and nobody acknowledges. The conversation continues as if nothing happened. But without saying a single word, the leaver has said “I do not need you” and “I do not agree with you.”

And now the others in the group must sit with this rejection. Whether the leaver’s silent verdict lands as relief or self-doubt, will tell you everything you need to know about the group.

There is a long-standing and rich body of work on non-verbal communication. Goffman studied how we perform for different audiences. Ekman mapped the micro-expressions we can’t suppress. What neither of them could have anticipated is that online communication would not just digitise the signal –  it would make the absence of a signal a signal in itself.

But back to my friend and the missed photo. I spent ten minutes on the phone explaining that I loved the photo, loved her daughter, loved the graduation, loved her. She said she knew that. She just wanted to check. After we ended our conversation, she sent me a 🙏🏼.

Don’t think too hard before you react to this newsletter.

Thank you to everyone who read my last post and wrote in. It feels so good to be back in touch with you. Reading your emails, I found that designers, writers and marketers all asked me some version of the same question: It’s not that bad, is it?

Designers, writers and marketers all asked me some version of the same question: It’s not that bad, is it?

Here is what I think, based on what I am actually observing on the ground.

The More Things Change…

The thing that will protect your career right now is something old and fundamental: the trust your client, your boss, or your colleague places in you. That trust is your real asset, and right now it is both more fragile and more valuable than it has ever been.

When my partner and I built Ray+Keshavan, we were operating in a market where branding was still regarded with suspicion by Indian business. Design was decoration. Brand strategy was a soft expense that finance directors tolerated and CEOs periodically questioned. What got us through board doors was a very specific kind of trust. Clients believed that we would deliver solutions that would work, that would endure, and that we could defend in the language of business rather than the language of aesthetics.

Clients knew that we could defend solutions in the language of business rather than the language of aesthetics.

There was this one instance where we had a standoff with the CEO of the country’s largest multinational FMCG company. He asked if we would put our money where our mouth is and agree to link part of our fee to an increase in sales. We could see this was an unfair situation for us: what happens down the funnel to the sales of a mass good was not something we could control. We also knew that this was an unprecedented offer and so we took it, knowing that it would bolstor our reputation as someone who shared responsibility for high-stakes decisions.

That requirement to be capable, to be accountable for your work has not gone away. But a second requirement sits alongside it, and this is the one most experienced practitioners are underestimating.

How To Be Trusted Today

Today, you must also be able to demonstrate that you have evaluated, experimented with, and formed a clear view on the tools that are reshaping how work gets done. Not because you are expected to be a technologist. But because you can only be trusted if you are not operating with a blind spot. A marketer or designer in 2026 who cannot speak with informed authority about where AI accelerates the work and where it introduces risk, is in the same position as a digital-era holdout in 2006 who thought the Internet was only for cat stories.

A marketer or designer in 2026 who cannot speak with informed authority about where AI accelerates the work and where it introduces risk, is in the same position as a digital-era holdout in 2006 who thought the Internet was only for cat stories.

On my current projects, I have been running every phase of the work through AI tools, partly to move faster and partly to understand where the tools hold up and where they fall apart.

AI is excellent at synthesis and iteration. Feeding it a hundred interview transcripts and asking it to surface recurring perceptions, contradictions, and gaps produces output that would have taken a researcher a couple of months. It also spots patterns that are genuinely insightful.

I am also about to test something I find very interesting: using AI to maintain message consistency across geographies without losing the strategic thread, then using it to flag drift as local teams produce their own content over time. The process involves layering translation tools, cultural prompting, and periodic AI-led audits against the original framework. This is still WIP and I will report back on what works and what doesn't.

The Answer, My Friend…

My answer to the question “It’s not so bad, is it?” is that it depends on you. On the one hand, the fact that a purely speculative doomsday article could move markets is in an indication that no one really knows what’s happening.

But for now, if we focus on what we can control, the marketers and designers who will come through this well are the ones who can sit in a room and say: I have tested these tools in my own work, here is where they change things significantly, here is where they require experienced oversight, here is the mistake we will make if we skip that step and here is where we just need to brainstorm and create, sans machines. That is a completely different conversation from “AI cannot replace human creativity,” which is what most creative folk are still saying, and which their audiences have largely stopped finding reassuring.

You cannot have that conversation if you have not done the work. The most urgent thing right now is not to defend the value of human judgment in the abstract. It is to go and build a specific, tested, honest point of view on what AI changes in your particular domain — and then be the person in the room who can communicate it with confidence.

Go and build a specific, tested, honest point of view on what AI changes in your particular domain — and then be the person in the room who can communicate it with confidence

That combination — the accountability that good practitioners have always offered, plus informed clarity about the tools that are changing how the work gets done — is what will earn you trust. Honestly, that has always been the job.

(This post uses em-dashes confidently)

Reading Y Combinator’s latest Request for Startups gave me a strong sense of deja vu. Among the ideas YC wants founders to pursue are “AI-native” agencies - businesses built with the assumption that agents, not humans, do most of the work.

We are sure to see more chatter around “I substituted my 10 person marketing team with an agent.” The subtext is clear: the more you automate with AI, the better your marketing outcomes automatically are. I am not so sure.

The subtext is clear: the more you automate with AI, the better your marketing outcomes automatically are. I am not so sure.

The ₹65 Crore Lesson

A few decades ago, there was a scramble for “digitally native” agencies and marketers.. Traditional agencies were declared dead and it was assumed that fluency with the digital medium would translate into marketing effectiveness. I remember a meeting with a company that had raised serious money and signed Hrithik Roshan, Shahrukh Khan and Sachin Tendulkar as brand ambassadors. Their entire pitch: they were building a financial portal. I asked some basic questions. What financial product were they actually selling? Who was the target customer? What was the revenue model? How were they different from existing financial services?

The client team did not appreciate my questions. They wanted to discuss websites and opening animations. They said “we were not digital-enough.”

Six months later, the company shut down. It turned out they had spent ₹65 crore on marketing a product nobody could quite define.

Over the years, digital has turned out to be another brand channel. A critical one, that has spawned a new generation of brands, but as has been proven over and over again, channel fluency is not a substitute for upstream foundational brand thinking. I believe we are heading toward a similar reckoning with AI, but this time the waters are murkier.

Fluency Without Fundamentals

The Judgment Gap, a recent Harvard Business Review article by David Duncan crystallised something I have been observing in my own work. Duncan, a strategy consultant, noticed that generative AI was helping him far more than it was helping his junior colleagues. He expected AI to help less-experienced workers punch above their weight. Instead, he watched the opposite: the tools amplified existing judgment rather than compensating for its absence.

This matches my recent experience.

The tools amplified existing judgment rather than compensating for its absence.

I am currently advising on a repositioning for a “hot” B2B technology startup. The internal marketing team is essentially focused on content creation. Their use of AI is limited to creating LinkedIn posts and event brochures quickly, with multiple options and ever-brighter gradients.

I have decided to use AI tools extensively throughout the project, partly to move faster, and partly to explore and document how AI can actually augment small, inexperienced teams.

Early in the project, I asked the team to develop interview guides for stakeholder research. What came back was polished and professionally formatted. Fifteen structured questions covering brand perception, competitive positioning, and satisfaction metrics.

It was also almost entirely useless.

The guide failed to account for several things that experience teaches you about interviewing C-suite executives.

For example, a CXO will not sit through fifteen structured questions - you will be lucky to get them talking freely for twenty minutes, and your job is essentially to guide a conversation so you gain insights from it.

Many of the questions had also drifted into customer satisfaction territory rather than staying focused on brand perception. These are different things.

An experienced marketer would have spotted these problems immediately. Not because they are smarter, but because they have the context and lived experience from dozens of these interviews. They have watched executives disengage from rigid questioning, learning to distinguish between what sounds important and what actually yields insight.

Without the judgment to evaluate the AI output, the team could not tell the difference between a slick-looking document and one that would actually work. The junior person had none of this context. The AI certainly did not provide it.

Without the judgment to evaluate the AI output, the team could not tell the difference between a slick-looking document and one that would actually work.

Hiding Under the Gloss

Before AI, a junior marketer creating an interview guide would have struggled visibly. The output would have been rough, incomplete, obviously in need of review. That roughness was a feature, not a bug – it signalled to everyone that more work and input was needed.

Now, AI produces output that look deceptively sophisticated and finished. It has the right structure, the professional tone, the appearance of completeness. This makes it harder to question, both for the person who created it and for the manager reviewing the output. The rough edges that once triggered scrutiny have been sanded away. The result is that errors travel further through the organisation before being caught. A flawed interview guide gets used. A positioning statement that sounds right but misses the market gets presented to leadership. Campaign copy that hits all the expected notes but says nothing distinctive gets published. The failure mode is not dramatic. It is insidious. Everything looks fine until it does not work.

Errors travel further through the organisation before being caught. The failure mode is not dramatic. It is insidious. Everything looks fine until it does not work.

More Dangerous Than Digitally-Native

The parallel that I made with digital is instructive, but it understates the risk. AI is different in three important ways.

First, AI tools can be used by anyone. The transition to the digital era required some technical fluency. AI requires only the ability to type a prompt. The barrier to producing professional-looking output has effectively disappeared.

Second, AI generates at scale. A junior marketer with AI can produce in a day what used to take a week. This sounds like productivity. But if the underlying thinking is flawed, you are now producing flawed work at scale. Volume creates its own drag on time and energy and makes it harder to pause and question.

Third, and most dangerous: AI creates a false sense of accomplishment. When you struggle to produce something, you know you struggled. When AI produces something polished in seconds, it feels like you have achieved something. The gap between effort and output has collapsed, and with it, the natural humility that once came from doing hard work badly before doing it well.

AI Augmentation Is Non-Negotiable

None of this is an argument against using AI. That ship has sailed. AI tools are transforming what is possible for small teams. A single marketer can now automate almost all their workflows: Produce competitive analyses, draft content at scale, synthesise research, and generate campaign variations at a pace that would have required a full team five years ago. The question is not whether to use AI, but how to use it without eroding the judgment that makes marketing actually work.

The question is not whether to use AI, but how to use it without eroding the judgment that makes marketing actually work.

The enthusiasm for AI-native companies and AI-native talent contains an implicit assumption: that AI capability is the scarce resource. This is a half-truth. What remains scarce is the judgment to direct AI toward the right problems, evaluate whether its outputs are actually good, and course-correct when they are not. That judgment is built through experience – experience that AI is now making it easier to skip.

For CEOs, the risk is not just that your marketing team fails to adopt AI. The risk is that they adopt it enthusiastically, produce work that looks increasingly professional, and gradually lose the ability to tell whether any of it is actually working. Twenty years ago, we learned that digital fluency was necessary but not sufficient. We are about to learn the same lesson about AI - and the lesson could be more expensive.

Meeta’s Note
I do not know yet what I will learn through this project about AI augmentation for non-technical brand-builders but I will try and share as much as I can. I am most excited about using AI to build institutional knowledge – to explore if this might mitigate the brand dilution that typically occurs with expansion and team turnover. For now, we addressed the flawed interview guide problem in the old-fashioned way: a brainstorming session with people from the sales and account management teams who had a deep understanding of the audience. Next up, an attempt to create virtual customer personas to test the guide with.

Hello

I’m going back to school. 

I start a fellowship at Stanford University next month, where I intend to focus on AI’s implication for the creative industry. I also plan to free-fall for some of the time to see where I land. 

What does this mean for THC?

That’s a question I’ve struggled with. 

Over these last few years, putting together the story and newsletter for you has been the highlight of my week. It’s seen me through highs and lows, pandemics and pivots. 

Realistically, I don’t think I’ll be able to do justice to the current THC format for the next year. Instead, I’ll try and share any interesting experiences from my time at Stanford through this Substack. (Please feel free to unsubscribe.) 

The THC site, with its repository of case studies and stories, will continue to grow. Our Instagram handle will also keep curating interesting creative work. 

I am so grateful that hundreds of brand-builders shared their stories with us and that thousands of readers like you gave us their time and attention. I hope that we were worth it.

Until we meet again, thank you. It has been a privilege to write for you. 

Meeta,

Founder, The Hard Copy

PS: I spoke to Deepak Gopalakrishnan and Utsav Mamoria about the THC journey for their excellent interview series, To Your Heart's Content (part of their program The 6% Club) Watch here.