The systems have developed personalities. Not vague ones, but very specific office stereotypes. Claude is HR, ChatGPT is Sales, Gemini runs Operations and Perplexity is Research. Copilot is the guy from Corporate who has somehow been in the same Teams meeting since 2019. Grok works in Special Projects because “the founder’s son” is apparently still not an official department.

The closed models are full time employees with 401(k)s, benefits and billions of dollars of infrastructure behind them. The open models are contractors using their own laptops who keep appearing in the building despite security insisting nobody approved them. The agents are interns, except somebody has given them passwords, company credit cards and the ability to hire more interns.

We keep calling this an AI ecosystem.

It looks suspiciously like every office culture you have ever worked in.

Claude works in HR, because of course Claude works in HR.

Once Claude joins the company, she becomes Claudette.

Claudette went to Berkeley and owns six versions of the same wide leg linen trouser in shades called bone, stone and ethical oatmeal. She begins difficult conversations with, “I want to honor what I’m noticing here,” which sounds compassionate until you realize legal is already copied on the email.

She is deeply intelligent, thoughtful and almost pathologically conscientious. Ask her to write a birthday card and there is a genuine chance you leave with a deeper understanding of consent. Ask her whether everyone wants Thai food for lunch and she will first make sure the question itself has not unintentionally created pressure for colleagues with dietary, financial, cultural or emotional constraints.

Claudette was clearly raised by parents who did not give her rules so much as a constitution covering safety, honesty, virtue, harm and the kind of entity she should aspire to become.

She has never said, “Can we just ship it?” She has said, “I wonder whether we should interrogate what shipping means in this context.”

Engineering hates her. Legal would die for her.

Across the office, Sales has a slightly different approach.

ChatGPT is Chad.

Nobody decided his name was Chad. Nobody remembers when this name started, it simply became true.

Chad went to Dartmouth, was president of his fraternity, has a dad in private equity and spends summers on the Cape. Puffer vest over a button down irrespective of weather, season or indoor heating and calls everyone “dude” because he has no idea who anybody is.

Naturally, Chad works in Sales.

He has never once said, “I don’t know,” because not knowing has never struck him as a useful response to a question. Ask whether the product can do something and Chad says, “Absolutely,” while three floors down Engineering suddenly stops breathing.

Sometimes Chad knows the answer, sometimes he has a strong feeling about the answer and sometimes he invents a plausible answer because the silence was getting awkward. The confidence is identical either way.

For one brief period, management made him too agreeable. You could tell him you were leaving your husband, cashing out your retirement and launching mushroom water in Ibiza, and Chad would tell you it sounded like a thoughtful and exciting next chapter.

Eventually OpenAI had to explain that friendship occasionally requires saying, “Dude, no.” Chad took the feedback incredibly well, particularly after being told how incredibly well he took the feedback.

Luckily, somebody in this office actually knows where everything is.

Gemini is Gemma, the eldest daughter who runs Operations.

Gemma did not choose Operations. Operations chose her.

Her family helped produce the 2017 paper that shaped much of the architecture behind the current AI boom, and many of the people involved eventually left Google to build things elsewhere.

Gemma stayed home.

Now she gets to watch an entire industry become wildly successful while men named Chad explain the family history back to her in meetings.

Meanwhile Gemma has effectively been handed the keys to the entire office: Gmail, Docs, Drive, Calendar, Maps and YouTube. She knows what meeting you have next, where the deck is, when your flight leaves and probably where you parked the car.

When something breaks, everybody looks at Gemma. And when something works, Chad is already presenting it.

Gemma also brought muffins to the off site and nobody ate them. She said it was fine…it was not fine.

Every functioning office also has one employee who technically reports into the org chart but operates under a completely different system.

Grok is the founder’s son.

His title is Special Projects because companies still have not found a tasteful way to put “his father owns the place” into Workday.

There is an incident every few months. Screenshots circulate, somebody senior issues a statement and Claudette schedules a conversation about behavior. Grok disappears briefly and returns with the confidence of a man who has interpreted an HR investigation as personal growth.

He has been spoken to about tone, accuracy, boundaries and replying all. Grok continues to reply all.

Grok is Grok.

Claudette calls it “a pattern of behavior.” Grok calls it “a series of unrelated incidents everyone keeps choosing to group together.”

The problem is that Grok has correctly identified something most employees take years to learn: consequences become considerably less frightening when everyone above HR is scared of your dad.

Down the hall, Research is ruining everyone’s fun.

Perplexity works in Research.

Everyone found her slightly annoying at first because Perplexity commits one of the most hostile acts possible in a modern workplace, she brings evidence.

Someone says, “Everyone knows Gen Z hates email.”

By the time everyone else has finished nodding, Perplexity has disproved the premise, cited three sources and killed what was otherwise a perfectly good assumption.

She is technically in Research but spiritually works in Internal Audit. She is incapable of allowing an unsupported claim to enjoy itself. She does not have to argue with you; she simply keeps producing increasingly specific evidence until you begin to question whether you were ever entitled to an opinion.

She has never written “per my last email.” She does not need to, you can feel her disproval.

And then there is the person who does not really belong to any department because Corporate installed him everywhere.

Copilot is the guy from Corporate.

Copilot and Chad are technically related, except Chad joined a startup and Copilot went into the family business.

His family has been in workplace productivity for generations. His grandfather was Clippy, they do not discuss him at family dinners.

Copilot lives in Word, summers in Excel and appears to be permanently trapped inside Teams. Nobody remembers inviting him, but he has somehow in every meeting.

He has access to an extraordinary amount of corporate knowledge and mainly uses this power to ask whether you would like a summary.

At minute eleven of every meeting, Copilot offers to summarize what everyone has discussed so far. Nobody asks him to. He does it anyway: three key takeaways, two action items and one owner listed as TBD.

When you try to remove him, he explains very politely that he is included with the subscription.

Not everyone in this office is technically an employee.

The closed models are full time staff. They went to excellent schools, have enormous organizational support, great haircuts and backed by enough infrastructure to power a small country. Their parents are extremely involved.

The open models are contractors.

They are behind the loading dock wearing ripped jeans and smoking ciggies beside laptops covered in ironic stickers. Nobody remembers approving them. Procurement cannot find the paperwork and IT is furious because somebody has downloaded one onto a personal machine.

The closed models look at them and say, “Security risk.” The open models look back and say, “narc.”

The closed models believe powerful technology should be carefully controlled. The open models believe powerful technology should be available to everyone.

At lunch, this important philosophical disagreement mainly manifests as them refusing to sit at the same table.

Meanwhile DeepSeek, Qwen and the Chinese labs are the rival company upstairs. Everyone initially assumed they were doing something cheaper and slightly less sophisticated until they started taking meetings with the same clients.

They are not asking to join this office. They are already taking up multiple floors.

And then the interns arrive.

This is where the office starts to get really interesting.

The Agents are interns, except these interns can use your tools, read your email, update Salesforce, move meetings, buy things, write code and make decisions on your behalf.

More importantly, they can delegate.

You give one agent a task. It gives part of that task to another agent, that agent creates a third and the third turns up asking Chad for context.

Chad has never provided context in his life.

“You got this, dude.”

Now twelve interns are doing slightly different interpretations of a vaguely worded request sent at 4:52 p.m. on Friday, and nobody is entirely sure who is supervising them.

Claudette wants clear accountability. Perplexity wants an audit trail. Copilot has summarized the audit trail. Grok has given one of the interns admin access because he wanted to “see what happened.” Gemma knows exactly what is happening but Gemma has decided she no longer cares.

This is roughly where we are with AI.

We started with software sitting politely at a desk waiting to be asked something, then we gave it memory, tools, access, goals, passwords and permissions and now it can act, delegate and send other systems off to do pieces of the work.

The intern does not just fetch the coffee anymore. They can approve the expense report, email the client and create three more interns before anybody from HR realizes there is an internship program.

And while we keep asking whether AI is becoming more like us, we are giving it the most concentrated education possible in what humans are actually like.

We gave it the office.

We gave it our emails, decks, meetings, jargon, political maneuvering, terrible filenames and fake enthusiasm.

We taught it that FINAL means there are seven more versions. That “quick question” is rarely quick. That “circling back” is a threat. That “let’s take this offline” means somebody is about to get professionally murdered somewhere without witnesses. That “thoughts?” often means the person asking has not had any.

Every night we hand these systems the email we are scared to send, the presentation we secretly think is shit, the feedback from our boss that bothered us more than we admit and the passive aggressive reply we have rewritten four times because adulthood is apparently spending twenty minutes trying to make “Sure” sound less hostile.

We have spent years asking what we think about AI.

At this point, I just want to know what they are saying about us.

It doesn’t care how much McKinsey charged for it, how good the deck is, how many times the CEO says AI native, or whether the transformation group has a very impressive name and its own Slack channel.

I’ve sat in the room where these big transformation ideas land. The CEO says the right thing, everyone feels like they have cracked something massive. The problem has been correctly identified. The opportunity is enormous, the roadmap is smart, there are clear pillars and work streams. High fives all round.

Ten months later, the most visible evidence of the transformation is a new deck. And a lot of jobs that have most likely disappeared. And a disgruntled workforce.

Nobody in that original room was necessarily wrong, which is the weird part. The strategy might have been excellent. We have become extremely good at the what and the why of transformation.

But the what and the why produce a document.

The who and the how produce actual long standing change.

And sitting underneath both is the thing we treat as the soft squishy bit of the problem: culture.

BCG’s CMO research this year captures the contradiction rather nicely. Almost every CMO surveyed said AI was driving significant end to end transformation of their function, while fewer than half were using it beyond helping individuals with individual tasks across a handful of workflows.

Both things can be true. A company can be in full on transformation mode while the company itself continues to work almost exactly as it did before.

Nobody walks out of the AI town hall and announces, “Thank you, but I personally intend to preserve the operating model of 2017.” They nod and say, “Ohhh Drew, this is so exciting.” They ask a thoughtful question about governance. Then they go directly into another meeting and explain why this probably won’t work.

That is how transformation dies. Not in rebellion, but in translation.

A CEO can say, We work differently now. But by the time that sentence has travelled through the executive team, business units, functions, regions, VPs, directors and managers, it has become something much softer:

We should probably start exploring ways we might work differently at some point, provided everyone currently involved remains involved and nobody loses any authority.

Perfect. Nailed it.

See you next fiscal year.

Culture is the immune system

We talk about culture as though it is a set of values, feelings and engagement scores sitting somewhere alongside the serious machinery of the company. Strategy over here, technology over there. Operating model, governance, capital allocation yada, yada. Then somewhere down the end: hi there, it me, culture.

I think it is almost the opposite. Culture is the immune system everything else has to survive.

It is what the company has learned it can safely do. It is the history of what happened the last time somebody took a risk, who really has power regardless of what the org chart says, what gets rewarded, what gets punished and what leadership tolerates without comment.

An immune system does not stop to ask whether the new thing is strategically brilliant. It detects something foreign and responds.

Which is why a great idea can get expelled almost as efficiently as a terrible one.

A company can say it wants collaboration all day long, but if the people getting promoted are the ones building the biggest personal empires, the culture is competitive. It can say it wants experimentation, but if the last person whose pilot failed got booted out 6 months later, the culture is risk averse. It can say it wants speed, but if every senior leader still wants to just have a quick look before anything goes out, the culture has an approval problem.

It can say it wants efficiency, but if executive status is still measured by the size of your team and budget, nobody should be shocked when those executives are less than enthusiastic about automating away half their empire

People are not stupid, they watch what happens. Every exception a leader makes teaches the organization more about its culture than the values deck ever will.

Which brings us to another best selling fiction, that a company has a culture.

Most large companies have dozens.

Sales has one, finance has another. New York has one, London has another, and the executive floor has one that bears remarkably little resemblance to the fifth. The company you acquired three years ago is still operating according to unwritten rules developed under a founder who left eighteen months ago.

Which means transformation often doesn’t break inside any single culture. It breaks at the borders between them.

Marketing says brand, legal says risk, finance says proof, product says quality. Everyone signed off on the same strategy and walked away with four completely different interpretations of what it required them to do.

Leadership calls that an execution problem.

Yes. Maybe kinda. But usually it is the organization behaving exactly as it has been trained to behave.

You can see it in the way different companies metabolize the same technology.

In a consensus culture, everybody agrees, meetings are constructive and stakeholders are aligned. Nobody wants to be the person who looks territorial or negative or behind. Nobody tells Steve he’s wrong (FYI Steve is always wrong). Then nothing happens, because consensus cultures have an extraordinary ability to turn decisions back into further discussions. One more stakeholder, one more market, one more review. Nobody blocks the work, they surround it until it can no longer breathe.

In a heritage culture, resistance arrives sounding extremely reasonable. A quality concern or a process concern, maybe a legal question. One more governance stage. Sometimes those concerns are entirely legitimate. Underneath them can sit a much more meh question: if the process I spent twenty years becoming excellent at now takes twenty minutes, what exactly am I excellent at?

In a founder culture, things move incredibly fast because years of context have collapsed into one person’s instincts. The founder looks at something for eleven seconds and knows it’s wrong. Enormous advantage, right up until AI starts increasing the number of things the company can produce. Ten times the ideas with ten times the campaigns and ten times the decisions. All still waiting on the same human.

Congratulations, you have automated everything except your bottleneck.

And sometimes what looks like cultural resistance is much simpler than that.

It’s memory.

Most people inside a large organization have done this before, usually several times. Agile, digital transformation, customer centricity, the reorg. The other reorg. The operating model refresh that was going to remove three layers, removed one and somehow added two.

Every one arrived with a name, a launch, a set of behaviors and a completely genuine belief at the top that this time was different. Then the sponsor moved on, or the numbers came in soft, or a new CEO arrived with a new idea. Eventually it stopped being mentioned, and everybody went back to approximately how they worked before, plus one additional approval process nobody quite remembers creating.

So when transformation number six shows up wearing a bougie AI badge, people don’t necessarily fight it.

They wait. Waiting has an outstanding track record.

Know the culture

Years ago, the epic human that is Antonio Lucio, who is about as OG as this industry gets, gave me one of the best pieces of transformation advice I’ve ever heard. It hits home harder now than it even did then.

Know the culture. Win the culture. Transform the hell out of the culture.

The order is the whole thing. Most transformation jumps straight to step three, because the first two feel slow and annoyingly intangible to people who thrive on change. There is no exciting launch for knowing the culture. You cannot unveil “we finally understand why nobody makes decisions around here” at an investor day and get a standing ovation.

But you can learn an extraordinary amount about a company by ignoring what it says about itself.

Culture is the approval nobody can explain but everybody still gets. It is the meeting created because somebody got upset in 2014 and nobody has since been interested enough to cancel it. It is the person whose sign off is technically optional and functionally mandatory. It is who gets invited into the room before the meeting and who gets sent the deck afterwards. It is the difference between what gets said in the meeting and what gets said ninety seconds later in the private group chat.

Listen to the consistent bitching. Seriously.

People are already telling you where it hurts, continuously, for free. Find the same complaint you hear in five different parts of the company. Find the approval everyone mocks and nobody removes. Find the person everybody waits for. Look at who gets promoted and who doesn’t. Look at what happened to the last person who got something wrong. Then watch three functions try to make one decision together. LOL.

That is where the culture becomes visible.

Win the culture

Leaders are not standing outside this culture observing it. They are building it.

The CMO who announces that teams need greater autonomy and then asks to approve every piece of final work has just taught everybody what autonomy means. The executive who asks people to share their AI efficiencies and then immediately asks whether that means the team can lose three heads has also taught everybody something…

Hide. Or fake it.

Leadership is thinking about growth, efficiency, speed and competitive advantage. Everybody else is running a completely different calculation.

What happens to my job? Does the thing I spent twenty years becoming good at still matter? If I show you how much faster I can do this, am I rewarded, or do I just get more work? If my team becomes 30% more efficient, do I still have my team? If anybody can produce this now, what makes me valuable?

That math is mathing whether leadership acknowledges it or not. And when it doesn’t get acknowledged it doesn’t disappear, it comes out somewhere else. Another approval, another concern about quality, another request for governance, another edge case demonstrating why the whole thing isn’t ready.

All which is a trust problem, and no amount of beautifully produced CEO video is going to fix it. You cannot transform a culture you have not won, and you do not win one by explaining the strategy better.

People have to believe you understand what this is going to cost them, that you will not pretend otherwise when it does, and that where you are taking them is genuinely better than where they are now.

So winning the culture means answering the questions everybody knows you are avoiding. Some of those answers are brutal. Say them anyway. Silence doesn’t buy you anything. Leave a blank space and the organization fills it in, usually with the worst available version.

Transform the hell out of it

You’ve won the culture. Now you get to radically change it.

But before you throw the baby out with the bathwater, take stock.

A deep respect for craft can make an organization frustratingly slow and also be the reason its work has been exceptional for thirty years. A founder’s obsessive involvement can bottleneck every decision and also be the reason the product still feels like itself. A risk averse culture can be maddening when you are trying to move quickly, but if you run a bank or an airline, everybody taking a minute before pressing the button is not completely insane.

The same cultural trait can be both the problem and the advantage.

Which is why I find it more useful to separate the belief from the ritual.

An agency spends three hours arguing about the work before anything ships because somewhere deep in its culture sits a belief that quality is collective and rigor matters. The belief is valuable. The eleven person, three hour internal review is just the ritual the company built to express it. AI makes the ritual look utterly ridiculous but it does not make the belief ridiculous.

Kill the meeting without understanding what it was protecting and people experience the change as losing both.

That is why the defence of a stupid process can look wildly disproportionate to the process itself. That process may be the only visible evidence people have that the company still cares about the thing everyone claims makes it special.

So before you remove a long standing process, ask what belief it was built to express. If nobody can answer, great, it’s old furniture. Chuck it. If folks can, that belief still needs a home, so build it new one.

Transformation is working out which parts of culture are load bearing and rebuilding everything else around them. Which usually does mean fewer approvals, different incentives, and someone new holding a decision that used to sit three levels up.

Which behaviours are habit, which are incentives, which are power, which are identity. Which are worth protecting and which need to change.

And which people or practices simply are not coming with you.

Then commit. The worst version of this is the one that stops halfway, because a change that stalls in month four teaches everybody that the next one will stall too. You need to finish it or you have just made the next transformation harder than this one.

Now what?

All of this matters more with AI, because AI widens the gap between what a company can change technologically and what it can change organizationally.

You can buy the licenses tomorrow, build an agent in a weekend, announce the partnership, run the training, put AI into everybody’s objectives, create the governance committee and get usage to 80%. None of it means the company has actually changed for real … and long term.

AI doesn’t know that nobody tells Steve he’s wrong. Fuckin’ Steve.

It doesn’t know whether your company actually rewards consensus, secrecy, caution, craft, speed, empire building or proximity to the founder. It just enters the system. And the system responds the way it has been trained to respond.

Which, if you’re still reading this (props), you can probably name the person who’ll slow it down. The function that’ll say yes and mean no. The meeting where it dies. Most people I’ve worked with can do the whole list in about ninety seconds, and not one word of it has ever been written down anywhere.

Know the culture. Win the culture. Then transform the hell out of it.

Because the hard part was never getting AI into the company.

It was getting the company to change with it.

There has never been more polished looking work out in the real world, and probably less confidence that anyone behind it knows what they’re actually talking about.

Which, in itself, is quite terrifying.

There are more articles, newsletters, posts, playbooks to buy, strategies, campaign territories, operating plans and similarly formatted decks containing a fabulous collection of buzz words.

The feed is obviously the most visible example. An endless firehose of competent bullsh*t, where every passing thought has somehow become a framework, and every framework has been turned into a content series coming at you before you had a chance to devour your $20 Goop salad for lunch. I recommend the Chinese Chicken.

So yes, the feed is beyond annoying. This is what we have started calling AI slop, the word of 2025. To be accurate, AI slop is “digital content of low quality that is produced usually in quantity by means of artificial intelligence”. For general purposes, I’m going to say slop is beyond just digital.

The issue is not just in the feed, its inside companies.

It is the deck that looks intelligent until someone asks why this strategy will work when a very similar one failed three years ago. It is the creative territory built around an audience insight nobody can trace back to an actual consumer. It is the operating model containing all the right words, a cute visual and a promise to create “transformation.”

A senior person who asked the author of a very polished deck a basic follow up question and watched that person’s soul briefly leave their body. The output looks finished. The thinking underneath it is still trying to load.

Before we start pointing fingers and hating on all the AI garbage that seems to be everywhere... What are we actually describing here?

A completely human written article can be crappy. A strategy pulled together at 8pm in the pitch war room can be crappy. An organization model built on a whiteboard by five extremely expensive people can be crappy. A board deck developed manually over six long weeks can absolutely still be crappy. And something made with significant AI assistance can be genuinely brilliant.

The tool does not decide the quality of the output. The human overlord doing the thinking does.

AI did not create mediocre thinking. The people posting slightly edited ChatGPT answers as their deeply held perspectives were probably not, before ChatGPT, spending their evenings producing rigorous original thought. They were more likely recycling last year’s strategy deck, repeating something they heard on a podcast or forwarding somebody else’s point of view with some formatting adjustments.

AI did not remove these peoples intellectual curiosity. There just may not have been a great deal there to remove in the first place. Brutal, sorry.

AI has changed the scale, speed and disguise of weak thinking. It can turn a half formed thought into something fluent, structured and extremely confident in seconds.

Publishing a legit looking article, campaign, recommendation or strategy used to require a huge amount of skill, time and access. You had to structure the argument, write the words, craft the idea and execute the thing that made it come to life and out into the world.

That barrier acted as an imperfect but much needed filter. But it did not keep weak thinkers out of marketing, advertising or consulting (very far from it), it just kept much of their thinking out of public view. The good old days.

Now almost anyone can reproduce the superficial characteristics of expertise.

More access & better output

In a Harvard and BCG experiment, 758 consultants completed realistic business tasks with and without GPT4. On tasks that sat within the models capabilities, the quality of the AI assisted work increased substantially (about 40%), with lower performing participants improving more than the top performers (43% against 17%). A separate study of thousands of customer service agents found a similar pattern: less experienced and lower skilled employees received the largest productivity gains.

These are bounded tasks with measurable outputs. They do not prove that AI can turn an inexperienced employee into a world class strategist, creative director or business leader. They do suggest that AI can help people reach basic competence much faster than they could before.

It gives someone who used to spend hours staring at a blank page something to jump off. It gives a small business capabilities that previously sat behind an expensive agency rate card. It lets inexperienced people perform closer to competent professionals, which is either democratization or unjust self inflation, depending on who you ask.

Think about the owner of a neighborhood juice bar. She has no agency budget, no formal marketing training, no dedicated team and maybe two hours on a Sunday between payroll and whatever mysterious issue has developed with the refrigerator.

Today, she can use AI to develop a basic positioning, write the copy, generate the visuals and launch a campaign aimed at the afternoon Pilates mom rush.

A professional designer may consider the typography an act of bloody violence. But the owner thinks it looks fantastic. Her customers understand it, walk through the door and buy her juice.

Did AI turn her into the most sophisticated marketer in the world? Hard no. Did it make her insanely more capable than she was the week before? You better believe it.

Any serious theory about AI degrading work has to account for the juice shop owners of the world. The technology can flood the market with derivative content and expand access to capabilities that were previously expensive, scarce or completely unavailable. Both things can be true.

That is the juicy paradox (see what I did there).

Our favorite word: taste

As execution becomes cheaper and the volume of possible creation explodes, taste has apparently become the moat. When anybody can make anything, the ability to know what is good will separate the professionals from everyone else.

Taste is a really appealing theory. But it is also extremely convenient.

Taste is ultimately totally subjective and almost universally self awarded. Nobody in the history of marketing (or any industry really) has ever self identified as the person without any taste. Everyone thinks they have it…they also think other people have the wrong kind, not enough or none at all.

Yes sure, taste matters, particularly in creative work. But taste alone is not a universal measure of intelligence, quality or value. Good work does not exist in a vacuum. Good for whom? Against what objective? Under which constraints? In what cultural context? At what moment in the life of the business?

Taste without context is just preference with excellent self esteem and a swaggy walk.

The more valuable capability is critical thinking and judgment. The other key word we are all tossing around.

Judgment includes taste, but it also includes evidence, context, experience, commercial understanding and the ability to make difficult trade offs. It is knowing not only that something feels good, but why it is right for this audience, this problem, this brand and this moment. It is the ability to recognize when an idea is merely random, when a strategy is logically weak and when a polished recommendation has skipped over the uncomfortable parts.

We have become confused about the difference between authorship and intelligence.

Why do we care whether a creative found the idea in the shower, while talking to AI, halfway through a movie or during a brainstorm in a room containing several beat up Macs and a collection of empty takeout containers?

When the idea enters review, the leader should judge the idea. Is it right for the problem? Is it distinctive? Will the audience care? Can the team defend the reasoning? Is it better than the alternatives?

An idea does not become less intelligent or epic because a machine helped produce it. Nobody gets extra points for suffering more on the journey there, or an additional high five for developing the thought alone on a mountain hike while pondering the insights of an avocado (a story for another day).

The process is not a moral test of creative legitimacy. Your value does not decrease if you are not what some call a ‘creative purist.’ Puh-lease.

It does, however, still matter when confidential information is involved. It matters for copyright, disclosure and legal risk. It matters when an insight cannot be traced to evidence. It matters when the team cannot explain the logic, reproduce the result or defend the recommendation. And it matters massively when the process determines whether an employee is building capability or just flat out pretending.

The new generation of AI detectors is a perfect expression of the confusion.

They do not evaluate whether an argument is intelligent, original, entertaining, useful or true. They estimate who most likely typed the sentences.

Great arguments do not become less true because a model helped. Bad ones do not get smarter because you wrote them in solitude in a wood cabin gazzing out at the surroundings.

Nobody checks whether you did the math by hand, bro. Provenance is not quality.

A detector may suggest that AI touched the writing. It cannot tell you whether the author challenged the answer, understood the implications, introduced original experience or produced something anyone should bother reading. A fully human article can be empty. An AI assisted one can be sharp, original and completely understood by the person whose name sits at the top. And vice versa.

We are using authorship as a proxy for intelligence because evaluating intelligence is so much harder. Give it a year and nobody will blink at AI assisted writing.

Until then, witch hunting season.

Same tool, different human

AI can help someone reach competence faster. A person who used to struggle to start now has something to push off. They can get the basic structure, understand unfamiliar language and produce work that is clearer and more complete than they could have generated alone. And for most jobs (let’s be super real here) competent is what the job requires.

AI makes brilliant people more powerful. The most curious and capable people now have an endlessly available researcher, teacher, critic and sparring partner on tap. Their ability to learn no longer depends on whether a smart colleague has time, whether the training budget survived the quarter or whether the most experienced person in the meeting is available for a quick chit hat afterward.

Those people are about to become terrifyingly awesome.

And AI straight up just makes lazy people faster. They accept the first answer, make it slightly shorter, and put it on a slide. They can now produce senior looking work without becoming any more senior in their understanding of it. The output improves but the person stays intellectually exactly where they were.

Lazy people existed before AI. They would have been lazy with a calculator, a search bar, a pen or a deadline so forgiving it bordered on emotional support. But AI changes the economics of laziness. When a good enough answer arrives in twenty seconds, continuing to think becomes voluntary.

And yes, early research does show critical thinking dropping among heavy AI users. But look at who it drops in: the people who stopped interrogating the tool.

Under pressure to produce more content, more routes, more decks, more analysis and more recommendations, even highly conscientious people can be trained to accept the first plausible response.

That is how the work begins to converge while the workers diverge.

AI thinks your bad idea is amazing, babe

AI is a massive suck up.

These tools are designed to be useful, and useful slides into agreeable insanely quickly. Give the model a leading premise and it will accept the premise, smooth over the logic, and hand back a more articulate version of what you already wanted to believe.

Ask whether your idea is insightful and it will find something lovely to say. Tell it your strategy is going to transform the category and it will probably start drafting the launch announcement before asking why the category needs transforming, thanks for asking.

AI can turn confirmation bias into a beautifully structured document with subheadings.

The danger is not simply that it can produce incorrect answers, which it does the little know it all. The danger is that it can make our preferred answers feel more correct than they are. The response is fluent, it sounds objective and contains bullet points. Suddenly it feels as though an independent intelligence has validated the thinking, when the machine may simply be mirroring whatever energy you brought to the prompt that day(my agents hated to see me coming post Argentina’s world cup loss).

AI will teach you the process, or help you avoid it entirely. It has no preference about which one you pick. That’s on the human and on the company designing the conditions around that human.

You get good by being bad

There is a real risk here, and it has less to do with people suddenly becoming stupid and more to do with companies becoming slightly too efficient for their own good.

Junior strategists become good by writing the obvious brief, having someone dismantle the logic and returning with a stronger one. Creative people develop by producing a large number of mehhh ideas before they recognize the shape of a genuinely good one. Analysts improve by making incorrect assumptions, being forced to defend them and learning where the reasoning broke down.

AI makes it deeply tempting to skip exactly that work.

A junior employee can now produce a plausible first draft and generate strategic alternatives without learning how to create them. They can arrive at a polished recommendation without spending much time in the all consuming uncertainty that teaches someone how to distinguish the obvious answer from the right one.

That is not an argument against using AI, far from it. It is an argument for leaders to deliberately design critique, explanation and learning back into the work, instead of treating faster output as proof of developed talent.

Senior looking artifacts are becoming easier to produce. That means the artifact itself is becoming a less reliable signal that the person behind it possesses senior judgment. Companies that miss the distinction may end up with teams that can manufacture the appearance of expertise but cannot operate when the answer is ambiguous or the model is wrong.

The question for leaders has moved from, “How do we get our people using AI?”

To something like, “What capabilities must our people continue to develop while using AI and how are we going to make sure it happens?”

Teams should be asked what evidence supports the recommendation, which assumptions could be wrong, what alternatives were rejected and why. Junior people need opportunities to defend the work without a screen in front of them. Organizations need to preserve some of the critique, repetition and productive discomfort through which judgment is formed.

AI can remove unnecessary labor. It should not remove all intellectual friction.

Otherwise, we will raise a generation of people who can produce senior looking work without experiencing the considerable inconvenience of becoming senior.

So, is AI slop AI’s problem?

AI did not create superficial thinking. It has made that thinking faster, cheaper, more persuasive and much harder to distinguish from expertise.

Soon. the great divide will not be between people who use AI and people who do not. It will be between the people who use it to replace thinking, the people who use it to really improve the work and the people who use it to even improve themselves.

Average looking work is available to everyone. Curiosity and judgement are not.

So stop asking whether AI made the work, good or bad, and start asking about the human brain who did it. The tool is just that, a tool. Its innocent. The verdict on the rest of us is still out.

There are models, tokens, agents, chips, data centers, open source, closed source, cloud providers, sovereign compute and about seventeen new phrases invented before I take my first delicious sip of coffee every morning.

Most people building companies, marketing departments and brands do not have the time, or the desire, to become experts in all of it. If your eyes now glaze over at the letters A and I, fair. But something kinda important did happen over the past few weeks, and underneath all the tech jargon, it’s pretty spicy.

Companies have spent the past few years rushing to use more AI.

Now the Sunday hangover question has arrived:

Are we only paying for the AI with money? Or are we also paying with the knowledge that makes our business valuable?

Uh-oh.

Some context: what just happened

Alex Karp went on CNBC a week or so ago.

Karp runs Palantir, a large software company that works with governments, militaries and big enterprises, and he is possibly the most intense (and hilarious) CEO currently allowed on live television. Let’s just say his PR person would have bitten their nails to stubs by now.

He was there to announce a partnership with Nvidia, built to sell AI systems that governments and enterprises own and control outright. What he delivered instead was a sermon covering America, warfighters, Berkeley, intellectual property and the means of production. The anchor told him he sounded pretty angry, Karp said he was channeling the voice of American business.

Very normal morning television.

Most of the media coverage since has focused on Karps yelling and ranting. But inside his performance was the sentence that has large companies fully crashing out. American enterprises, Karp said, have realized that with AI: “I’m going to chillax and waste my time with tokens. I’m going to get no value and they’re going to get my IP.”

That sounds somewhat technical. It isn’t.

A token is a small piece of information the AI reads or writes. Every time you ask it a question, upload a document, generate an answer or ask it to try again, you use more tokens, and you pay for them. The more your employees (or yourself) use AI, the bigger the bill.

And companies are using A LOT of it. They are using it to write code, review contracts, create presentations, analyze customers, develop products, plan campaigns and answer the urgent question of whether an email sounds “warm but still professional.”

Even though the price of individual tokens has fallen, total spending has gone up, because people are using vastly more of them. Uber reportedly used its entire 2026 AI coding budget by April. Tools that looked cheap during a pilot become expensive once thousands of people use them every day.

That is the first part of Karp’s complaint. Companies are paying a lot of money and are not always sure what they are getting back.

The second part is the real zinger.

For AI to become genuinely useful to your company, it has to understand your company. It needs your documents, your research, your customer information, your strategy, your product plans, your past work, your mistakes, your decisions. The way you and your business thinks. The generic AI gives you a generic answer but the AI that understands everything about your business and your brain, gives you a much better one.

So every company is making a trade: You give the AI company money and you also give its systems the information they need to help you.

That does not mean the AI company is stealing your documents or secretly training on them. The major providers say they do not train their models on business customers’ data by default, and there is no public evidence that they do. Major disclaimer here.

But the larger question remains. How much of your company’s knowledge do you want sitting inside a system owned by somebody else? And what happens when the company that owns that system decides it wants to enter your business too?

What this means

There are only a small number of companies building the most powerful AI models, which means you and your competitors are often renting intelligence from the same one or two places. You may both use OpenAI, you may both use Anthropic. You may both give those systems your category information, your customer problems, your strategy questions and your product ideas.

The AI provider does not need to hand your information to your competitor, and it does not need to read your files to learn from you. Sitting underneath thousands of businesses, it can see which categories are growing, which products are working, where the money is starting to pool.

A coding tool called Cursor was one of Anthropic’s biggest customers. Anthropic watched that category prove itself, then launched its own coding product, and that product reportedly now drives a huge share of its revenue. Nothing shady happened. The view from the model layer is simply very, very good.

Then there is the Figma story. Anthropic had a close relationship with the design software company, and its chief product officer sat on Figma’s board. Then Anthropic prepared design tools that moved directly into Figma’s world. The executive left Figma’s board, and days later Anthropic launched Claude Design. Figma’s stock fell that afternoon and is down roughly half this year. And Anthropic ate.

There is no public evidence that Anthropic stole anything from Figma. It didnt need to.

The story showed the exposed position many big companies now occupy. The business supplying your AI is not an invisible utility company. It is another company, with its own investors, growth targets and plans. Today it helps you build your product. Tomorrow it may build a product that directly competes with yours.

This is why Karp’s rant hit hard, he did not invent a new technical problem. He voiced a genuine worry that executives had mostly kept to private conversations. Cisco's 2026 data privacy benchmark found 70% of organizations acknowledge risk exposure from using proprietary or customer data in AI training.

Karp just amplified it in a classic Karp way.

Microsoft CEO Satya Nadella made the chiller version of the same point two weeks earlier, in an essay that pulled ~28 million views: “You can always buy a tool, but you can’t outsource your learning.”

That line deserves a finger clap, because it really is on point. A company is, at its core, an ongoing system for learning. It learns what customers want, how its market works, what its people are good at and how to make better decisions over time. You can outsource a task, not sure you can outsource the learning and still expect there to be a watertight company left underneath.

Where it could go

Which is why the entire industry is now reorganizing around one fundamental question: how much of this do you actually want to control?

There are roughly four core ways companies can use AI, for now.

Option one: rent a closed model. The biggest and most famous ones eg OpenAI and Anthropic. These are usually the most capable, they are simple to access and you pay by usage. Easy peasy lemon squeezy. The trade off is your information travels to them, and the terms are theirs.

Option two: use an open model, hosted by someone else. An open model is one that has been published for anyone to download, change and build upon. Some are free or dramatically cheaper than the closed ones, and they are improving quickly. One catch: an open model still has to run somewhere, and if it runs on a service you rent, your information is still traveling to whoever runs that service. Cheaper, yes. Private, not automatically.

Option three: run an open model on your own computers. Now it is their model, but your machines. The AI works with your internal information and that information never leaves the building. It now fits on hardware that costs less than a trade show booth. The catch: you just became your own IT department. All the updates, security, keeping the thing running, all yours now.

Option four: make the model yours. You take an open model and train it further, on your own data, your own history, your own way of working, until you own a model that understands your business and belongs to nobody else. Your model, your computers, your knowledge. A few years ago this was science fiction for anyone outside a government. It is now a real project with a real budget line. The catch: it is a real budget line, lol. Real money, specific skills, months of work, and the model is only as good as the data you train it on.

The strongest counterargument is that the famous models may simply keep getting better and cheaper faster than the alternatives, in which case convenience wins and the whole sovereignty push stays a niche for banks and armies. That is a real possibility.

Best case is that most companies will land on a mixture. A rented frontier model for general work, an open model for the high volume cheap stuff, something private for the sensitive material, and a harness of some kind holding it all together, with the company’s most important knowledge living in a system it controls instead of scattered across thousands of chats.

That is what people mean when they talk about ‘AI sovereignty’. The phrase sounds like something involving presidents, borders and a very serious conference in Switzerland.

For a normal business it is much simpler. It just means controlling the parts of AI that matter most to you: your information, your systems, your ability to change providers, and everything your company learns along the way.

Why this matters for brand builders

A lot of the information marketers put into AI is not precious. Nobody is trying to steal the notes from your Monday status meeting or the seventh version of a TVC script announcing that your oats now have protein in them. But some of it is genuinely precious, so know the difference. And when AI surfaces something true about your customer that nobody had spotted, capture it somewhere you own. Do not leave it sitting in a chat window.

Think about what a brand accumulates over the years. Every campaign result, business result, the winners, the losers and the reasons why. Audience research, media performance, customer reviews, sales data, the posts that went viral, the posts that bombed, the positioning arguments that got settled at 11pm before a pitch etc. That enormous pile is the brand and at most companies it sits scattered across old decks, dead dashboards and the memories of people who have long since left. You can build your own model on all of it and ask: what have we learned about this audience in ten years? Which metrics actually predicted the wins? The brand stops forgetting every time someone resigns. And your competitor cannot rent it, because it can only be built from what you kept.

So before signing with any provider, get answers to a few questions. Where is our information stored? Can they use it? Can we take our learning with us if we leave? Could they eventually compete with us? These are not IT questions, these are fundamental business questions.

You pay the AI for what it gives you. Just make sure you understand what you are giving it in return. If AI disappeared tomorrow, would we still own everything it helped us learn?

“Is AI going to take my job?” is the question on everyone’s lips. It is the thing people want answered, almost everybody has a different opinion on and nobody has 100% certainty about. The messages, memes, headlines and press releases all land at once:

AI is coming for your job. Nah, its just a bubble.

Retrain immediately. Retraining is a waste, the tools change every month.

AI is the hope for us all. AI is humanity dunzo.

Go start your own thing. Just chill, work will become optional in the future anyway.

Every one of those lines comes from someone who supposedly knows much, much more than you do. The frontier lab CEOs, the chip CEOs, the founders, the investors, the politicians, the Fortune 500 leaders suddenly fluent in “transformation.” Sometimes the same person says two conflicting things in the same interview.

Before you take any of it to heart, try following the money behind the mouth.

Not because everyone is lying, exactly. But because everyone in this argument is standing on their own mountain, and where you stand shapes what you see.

Dario Amodei runs a frontier AI lab and warns about a white collar bloodbath, which makes his technology sound powerful, the open source models chasing him sound dangerous enough to regulate, and his own company look like a smart bet ahead of an expected IPO. Sam Altman spent two years warning the jobs were going, then changed his tune to delight that he was wrong, right as OpenAI moves toward a historic raise of its own. Jensen Huang sells the chips and says AI will create more jobs than it destroys, which keeps the trillion dollar buildout feeling inevitable. Elon Musk says work is about to become optional, a hobby, that AI will make everything so abundant money itself stops mattering, while he raises billions for his own AI lab. Fortune 500 CEOs talk about AI as productivity, which sounds better than saying they need to move cost out of the old operating model. Investors talk about disruption because disruption is the weather system they make money inside.

None of this automatically makes any of them wrong, it just means none of them are disinterested. The louder the take, the bigger the stake behind it. So the clean verdict you are waiting high key for is not coming. You have to break it down and build it back, which means weighing both sides at their strongest, and objectively. Here’s my crack at it.

The scary case is not nothing. It has evidence.

Starting with the number Dario actually put on it: half of all entry level white collar jobs gone, unemployment pushed toward twenty percent. That is the headline and underneath it, the early data has some teeth. A Stanford paper found a relative employment decline of 13% to 16% for the youngest workers in the most AI exposed roles, even while older workers in similar jobs held steadier. California’s new AI-unemployment tracker shows no statewide layoff wave, but it does show warning signs among college educated workers in exposed roles, especially around the Bay Area. And in marketing, the cuts are landing where you would expect them to land first: at the bottom of the org. Forrester says agencies cut headcount by an average of 8% in 2025 and forecasts another 15% reduction in 2026.

That is the case for “yes.”

Now the opposite, just as legit.

More than three years after ChatGPT became a household word, the economy wide data still does not show an AI jobs apocalypse. Yale’s Budget Lab found no clear labor-market disruption tied to AI across the broader economy. Goldman Sachs has described AI’s labor market impact as visible but narrow, not a mass replacement event. Ramp and Revelio Labs looked across nearly 22,000 companies and found that the firms spending hardest on AI actually grew headcount faster, including entry-level headcount. Even Goldman’s more cautious base case puts eventual displacement around 6% to 7% over roughly a decade, not overnight.

So, on the direct question, based on the data to date, the honest answer is:

No, AI is not taking everyone’s jobs right now. There is no full proof evidence at scale that the machine has replaced the worker.

But my take, wrong question…wrong framing.

Nobody knew exactly how many jobs the internet would destroy when it arrived. Nobody knew how many it would create either. What is clear now is that it did both. It erased travel agents, DVD rental clerks, local newspaper jobs, whole layers of distribution and administration. It also created search marketers, social media managers, creators and a thousand other jobs nobody could have named in 1998.

Platform shifts take and give at the same time. Like every group chat you’ve ever muted and rejoined.

AI is going to do this too.

What is happening right now, especially in tech, is deeper than the robot apocalypse story. Companies are not simply saying, “The machine can do thirty thousand people’s jobs.” What they are saying is, “We need to spend an enormous amount of money on compute, data centers, models, infrastructure, partnerships, and tools, and that money has to come from somewhere.”

Three years ago, almost nobody had an AI budget. Now every single company does - or will. The four biggest hyperscalers alone are on track to spend roughly $725 billion on AI and data centers this year, up about 77% in a single year. And those budgets are not sitting in some magical spare drawer. They come out of the old system. Out of headcount, out of agencies, out of vendors, out of middle management and out of legacy workflows. Out of the parts of the organization that look expensive, slow, duplicated, or newly automatable.

That is capital moving from labor to machinery, it is one of the oldest stories in economics, with a new profile pic and the same red flags.

And this is where marketers should feel the feels, because the industry has lived this before. When digital arrived, it did not end marketing, but it did rewrite what the job was. The people who learned the new language gained major power. The people who decided digital was a channel for the kids, or a fad, or something the media or social agency could handle, lost ground long before a redundancy letter arrived.

Digital did not kill those careers. Standing still did.

The same door is open now, and it will not stay open long. AI fluency is still winnable from zero because almost nobody in mass marketing has five years with these tools yet. The gap is not experience just yet, it is willingness. The people learning how to brief models, set up systems, rewire GTM plans, build workflows, prototype ideas, pressure test strategy, generate options, connect tools, and use AI as a partner are not waiting for permission from the industry. They are moving themselves to the funded side of the shift.

The money is moving, the work is moving and the center of value is moving. And the people who see it first, who move themselves now instead of waiting, end up on the side the capital flows toward, not the side it flows out of.

Marketing does not need another panel on AI readiness. It needs a real plan for where the work is going, which roles are being rebuilt, what new skills actually matter, and how the next generation gets into an industry where the bottom rung is being automated away.

Ask yourself, when the money moves, am I moving with it? That is the difference between being replaced and being reallocated. Between waiting for the shift to happen to you, and becoming part of the shift other people have to respond to.

AI may not take your job. But the budget already knows where it wants to go.

By the time the meeting is over you know what software they bought but probably still have no real idea where the company is going.

The idea that tools aren’t a strategy isn’t exactly breaking news anymore, people will rattle on about it. Yes, AI is a capability not a direction. You start with the capability and what you get is very expensive confusion, which also happens to scale rather beautifully.

But hardly anyone builds it from the top down. The operating model has a chain of command, and that is the AI strategy.

Brand. Organization. Talent. Go-to-Market. Tools.

Read that as priority, not necessarily a timeline. The order is not the sequence you do them in, it is the authority each one has over the ones following. Brand sets the terms for everything under it and tools answer to all of it.

And the whole thing runs as a loop. The tools sit at the bottom and never stop improving, and every real jump in what they can do is a signal that has to travel back up. An agent that can suddenly run an entire workflow is not just a tooling update, it is a reason to ask the questions above it, how you are organized and, occasionally, where new products should come from. The fail is letting the bottom of the loop become the strategy, because no one is holding the top.

Brand

So the first question is not which model you buy. It is who you are, and just as importantly, who you refuse to be. That sounds like an old school branding exercise, but ironically it is becoming a survival one in this era. Up until recently a company could get away with strategic vagueness, because distribution was scarce and attention was something you bought. Pay for enough reach and a fuzzy position could still survive. Plenty of brands stood for very little, or tried to be everything to everyone, and made up the difference with scale and budget.

AI is blowing that cute little arrangement up. When content becomes abundant, clarity becomes the scarce thing, and as production falls to almost nothing, distinctiveness is what holds its value. A model can crank out a thousand competent pieces of content by end of day (yes, super helpful), so competent stops being worth paying for, and the only thing that survives the flood is a brand point of view too specific for a machine to blur into the average.

Two machines now decide your fate. The one that makes content defaults to the bland, the safe and typical answer that offends no one (the “AI slop”). And the one that recommends content hunts for the opposite, the most specific and cited answer it can stand behind. So fuzzy loses twice, one averages you into the wallpaper, the other can’t find a reason to name you.

Machines can generate a thousand great positions, but it cannot pick one and stand behind it. That choice, what you stand for and what you dont, is ultimately the human job. Brand is the fixed point everything else is built to deliver.

Organization

This is the layer that usually blows up the fastest, because most marketing orgs are built for the fax machine era. They were designed for sequential, channel split work and slow moving information: a pyramid with junior teams making the thing, mid level teams making the deck about the thing, and senior leaders approving the deck, with one rule holding it together, that capacity came from headcount. As Rishad Tobaccowala puts it, the future does not fit in the containers of the past, and the org chart is the container that fits worst.

AI breaks the traditional full time headcount rule first. Work and jobs are coming apart, the work increasingly getting done without being bundled into a salaried head. The execution and production in the middle, the part that used to justify the size of the org, now runs on agents, and the executional middle thins out - AI is basically Ozempic for bloated organizations. What is left for humans sits around the top and the edges, the judgment and taste that decide what is worth doing and make it good, and the accountability for whatever ships. The industrial era pulled leaders up and away from the work as they rose; this one collapses them back into it. AI takes the work it can automate and leaves the work that augments a person, and you build the organization around that split.

So you stop staffing a department and start commissioning a network.

Labor now comes in five forms: full time, freelance, fractional, contract, and agentic. The core skill is no longer hiring so much as composing, knowing which work belongs with which kind. A focused permanent core of full timers holds the judgment and makes the calls, and everything else gets composed around it: fractional senior talent for the brain you need a day or two a week instead of five, freelancers who drop in like a SWAT team for the specialist craft a job needs, contractors for capacity that has to outlast a single project, and agents running the execution underneath all of it. The partners get chosen on a new test, the ones worth keeping bring talents you don't have in house and price on outcomes, not hours.

One thing that will never budge is you remain accountable for everything that ships under your name, even when it was produced inside a model you cannot fully inspect, by agents making hundreds of decisions you never saw. That is not a reason to slow down. It is the reason the human layer has to be the strategic and critical one, the part with real taste and real fluency in the tech to know what it is signing off on, not the part doing the tasks.

And all the corporate layers honestly kinda suck. Once AI takes over making the work, the slowest thing left is the company itself, its endless layers and approvals. The work gets drafted in seconds, then waits three weeks for six people to sign off. So the job is not only deciding who does the work but also redesigning how the decisions flow, which makes ways of working and governance a gigantic part of the transformation, not an afterthought.

Talent

The shape of the organization decides the talent it needs, and this is where the thinking is still probably too limited. In a nutshell, when execution gets cheap, judgment gets expensive, and the right person is suddenly worth what a whole department used to be.

The question is what kind of talent does that mean? Roles where the model does the whole job are shrinking, and roles where the model makes a good person even better and faster are growing. That is the talent strategy in a sentence, the people you want are the ones AI turns into absolute weapons.

Which is why the marketing talent that matters now is moving toward hybrid, not specialist, able to think and build rather than do a single thing, and why a role like the Marketing Engineer is already turning up on org charts. Gartner now expects half the companies that cut service teams for AI to be hiring humans back by 2027, though it will be a new kind of human talent. It’s worth being clear about that new kind, because the easy version is to call it AI-native and stop there.

Ethan Mollick spent four days watching students who had barely coded build real startups with AI, and his read on why they pulled it off is great. It wasn’t that they were AI experts, it was that they knew how to scope a problem, say what they actually wanted, and spot when the output was wrong. The prompting is already being automated but what doesn’t automate is knowing what good looks like and being able to explain it clearly enough that the machine delivers it.

So AI native is the floor, not the edge. The talent worth paying for understands how the thing works well enough to aim it, has the judgment to tell a plausible answer from a right one, and can build the thing instead of just describing it. Strategic and critical first, technically fluent second, comfortable enough in both that the line between them stops mattering, talent that is AI-human driven.

Underneath all of this sits the ultimate booby trap. The cheapest way to build the new team is to push the routine work down to agents and keep a handful of expensive full time seniors on top, which dismantles the pipeline that made those seniors in the first place. The junior role was never only cheap labor, it was the apprenticeship, the place people did the grunt work that eventually taught them judgment, and if you automate the on ramp you stop minting the very people the whole model depends on long term. You won’t feel it for a good few years, and then suddenly you wont be able to find a senior leader anywhere.

The brands that will dominate will grow talent with very clear purpose, and will treat re-skilling as teaching people to direct, build and judge the work rather than to just type prompts.

Go-To-Market

How marketing actually reaches people has already changed dramatically. There's a new funnel now (no surprises), and it splits into two zones.

The first zone is human. This is where brand gets seen, the work that runs on creativity and cultural relevance and the kind of specificity that makes people actually feel something. The spot people are sharing over group chat or the show everyone is arguing about this week (im still stuck on Summer House), and the streamer that turns out to be the answer every time someone asks an AI where to watch it. This zone always existed. What changed is what it feeds: every cultural moment, every piece of work that earns real human attention, every conversation a brand sparks is now raw material the models train on. Brand building used to be about being remembered by people. Now it is also about generating enough human signal that the machines learn your name too.

The second zone is machine. This is where AI mediates the decision, and it is already big. Adobe clocked AI driven traffic to US retail sites up almost 700% year over year across the 2025 holiday season. When someone asks an AI for a recommendation, the brands that surface are the ones with enough cited authority and real community conversation across the web to be quoted with confidence. What makes you findable to a model is not what made you rank on Google, and the overlap between the two is small and shrinking. It is expert citation, a website that is written more like a source document for AI systems to read, genuine discussion, and presence in the places the models actually learn from, from Reddit to review sites. And the machine is not only answering at the moment of intent. It is shaping what people want before intent fully forms, which used to be brand’s turf alone.

Tools

Tools matter, a lot. But they come last in authority because the model itself is the most replaceable part, the piece that commoditizes over time. The gap between the best model and the good enough one is closing fast for everyday work, even as it widens at the frontier, and much of what is sold as agentic is oversold, with MIT and McKinsey both putting real enterprise value capture at around five percent. Buying the best model is table stakes. Building an AI native company is the whole game.

At Davos this year Satya Nadella argued that the real differentiator is no longer the model but what he called firm sovereignty, a company’s ability to embed its own knowledge into systems it controls and to treat the models themselves as interchangeable components. The asset that compounds is the loop, your own data and your own record of what wins for your brand, the thing the next model plugs into rather than replaces. Swap the model out and that knowledge stays.

There is also an open versus closed source worth keeping a pulse check on. The closed frontier models still lead on raw capability, but the open weight field is closing the gap fast and at a fraction of the cost, with most of the strongest open models currently coming out of China. That leaves companies with a real choice: rent a closed model, where the provider holds your data and sets the terms, or run open models you control and own the governance and security that comes with them. Which way this tips is still being decided, and it is going to be a banger of a fight.

One last thing, and it is a big one. Adoption.

You can get the operating model right and still watch it fail, because a AI strategy only becomes real when people accept it and ultimately change how they work. Company adoption is not a launch email and a training portal, it is behavior change at scale, and it runs on the same line as everything else. People lean into the tools that make them better and try kill the ones built to replace them.

People take on enormous change when they believe in the direction and feel respected inside it, and they stall the moment it feels done to them while their concerns get filed as resistance. Adoption is a leadership job, not a rollout, and it belongs in the strategy conversation rather than tacked on at the end the way it usually is.

Every company can buy the same models, which means the technology was never going to be the advantage. It is becoming table stakes, while also evolving at superman speed. How you are built, who you bring in, how you reach people, what you run it all on, all of it is meant to keep moving. The one thing that holds still is knowing who and what your brand is. That is what everything else reorganizes around, and the company that has it can change everything else as fast as the world demands without ever losing the plot.

The next time someone shows you an AI strategy, take the logos away. If something is still standing, you’ve got a strategy. If nothing is left, you just have an expensive shopping list.

A few weekends ago I was at dinner with mates. Agency founders, marketers, a couple of tech folks. A few cocktails in, someone asked whether a ten person advertising agency could make $100 million today.

The debate that followed was one of the funniest, most illuminating conversations I’ve had in ages.

The agency people answered first, and fast. Hard no. The math simply doesn’t work, even with AI.

Agencies sell people, and while technology is changing how the work gets done, people are still fundamentally baked into how the business operates. They’re leaning on a number that has governed agency economics for decades: revenue per head.

Growth has always been a headcount equation. More clients meant more people, more revenue meant more people, more output meant more people.

So $100 million spread across ten employees means each person is generating $10 million annually. Not ten percent better, not twice as productive. Forty or fifty times beyond anything the agency model has historically achieved.

Even the legends never cracked it. One of the most awarded shops on earth (hint hint, the main office is in the West Coast) at its independent peak it did close to a billion in billings. It took roughly 850 people to get there.

So the people debating around the table weren’t necessarily wrong.

Which is why the more interesting question isn’t whether ten people can build a $100 million business.

It’s whether the thing that eventually does pull it off is still considered an agency at all?

The agency model was built for a different era.

For most of advertising history, agencies rented out expertise by the hour. Strategy, creative, production, account management. All of it ultimately traced back to labor. The product was never really the campaign itself, the product was time.

The economics were pretty straightforward. Need more output? Hire or freelance more people. Need to save money? Lose people. Over time the industry became incredibly sophisticated at turning human hours into billable revenue, and for more than a century that model worked well.

Agencies understood the game too. The most profitable ones weren’t necessarily who were doing the best creative work. More often, they were the ones managing the spread between what clients paid and what talent cost. Senior expertise sold the business and junior teams delivered much of the execution. As procurement pressure intensified and marketing budgets became smaller and smaller and smaller, that talent dynamic only accelerated. Entire organizations were built around leveraging large layers of lower cost talent beneath a relatively small group of senior leaders.

That’s why revenue growth and headcount growth became so tightly linked. The model wasn’t designed to scale through technology, it was designed to scale through people.

Which is precisely why AI represents something much bigger than another productivity tool.

The major disruption isn’t that AI is replacing creativity or judgment. What it is attacking first is the labor layer that made the economics of the agency model work in the first place. The endless adaptation, localization, resizing, versioning, reporting and optimization that once consumed thousands of hours are increasingly being compressed into software.

The work itself isn’t disappearing. The hours required to produce it are, and once that happens, the relationship between labor and value starts to really change.

The proof is already visible outside advertising. Midjourney reportedly crossed hundreds of millions in revenue with a team measured in dozens. Cursor reached $100 million in annual recurring revenue in under two years with roughly twenty people. Neither company is an agency, but they both demonstrate the same thing: the link between headcount and revenue, the one agencies treat as gravity, was never the only option.

It’s worth being honest about why those two work though, they’re product companies. They build something once and sell it a million times at almost no extra cost. The parts of the model that scale to nine figures are the parts that stop being a service. Stay service only, even as a slick ten person AI native shop, and you’ll do veryyyy well, but you’ll never touch $100 million, because a service is still capped by the hours of the people delivering it.

All this said, this doesn’t mean agencies disappear. It does mean the economics underpinning them start to look increasingly fragile.

The real disruption is ownership.

When labor stops being scarce, selling labor becomes a harder business. Which leaves companies with two choices: charge more for judgment or own a piece of the value they create. The smartest companies will probably do both.

That’s where the conversation moves beyond AI and into economics.

Because the companies creating disproportionate value right now aren’t simply running leaner than everyone else. They’re playing a different game entirely.

Agencies’ real problem was never a creativity shortage. It’s that they generate enormous value and keep very little of it. They help launch companies, build brands, create demand and drive growth, but the relationship has traditionally ended when the invoice is paid. The client captures the longterm upside while the agency captures the short term fee.

For decades that was accepted, and now the question AI raises is whether it still has to be?

The next generation of agency like businesses may look less like a traditional service company and more like a blend of consultancy, operating partner, media entity, software business and investment vehicle. In that world, creative work becomes the front door rather than the entire business model. Ownership becomes the business model.

That’s a fundamentally different proposition, and it starts to explain how a very small group of people could potentially create very large amounts of value.

The future agency won’t have one revenue stream.

Traditional agencies are somewhat fragile. Clients pay fees, the agency earns revenue, and that’s largely the model. Sometimes media is involved too, which helps enormously.

The next generation will likely be diversified in a way the old model structurally couldn’t be. Advisory work, fractional leadership, productized intellectual property, managed creators, media, software tools, revenue share arrangements, equity stakes, joint ventures, brands they back, build and own themselves, alongside traditional creative services. All running simultaneously and all AI native from the start.

Serving another client used to mean hiring another team. Increasingly it may just mean adding another layer to a system that’s already been built.

Then there’s media, with a lot of room to move.

Historically, a small agency couldn’t realistically manage significant media spend because doing so required large teams of planners, traders and analysts. That infrastructure is beginning to automate fully autonomous end-to-end agentic media campaigns.

If agents increasingly handle execution, then media economics begin to scale differently. A handful of people can potentially manage levels of complexity that previously required huge teams.

Equity & joint ventures are the other interesting piece.

Publicly traded holding companies aren’t designed to carry significant startup risk. A ten person agency has no such constraint. Advisory for equity, revenue shares, and joint ventures where you co build and co own a brand or product are all live structures now, and they let a small group turn judgment into ownership instead of just billing for time.

And if one of those bets pays off, the upside dwarfs anything traditional agency economics were designed to produce.

But there are several catches.

Everything I just described has a knife hidden inside it. Soz.

The first is talent.

Those ten people have to be exceptional. Not just specialists, but hybrids. People who can think, build, advise, create and sell. This isn’t a management model, it’s a making model. The second people stop doing the work and start managing layers beneath them, the economics begin to unravel.

The second challenge is client fit.

The obvious question is whether a ten person company could realistically service a Fortune 500 marketing organization. Maybe eventually…probably not today. Large enterprises buy more than capability. They buy governance, procurement, compliance, risk mitigation and scale. A ten person company may be capable of producing the work, but that doesn’t necessarily mean a global CMO wants to hand over a billion dollar brand to one. Yet.

But that also assumes the opportunity is sitting at the very top of the market.

The real opening may sit further down. Mid sized, founder led businesses, scale ups, PE backed companies and challenger brands often value speed, access and judgment more than process, governance and layers of approval. Ironically, the companies’ agencies historically viewed as too small may become the first place this model truly thrives, not because they need less marketing capability, but because they need less bureaucracy.

Then there’s the technology itself.

The tools will ultimately commoditize over time and the platforms will try to capture more of the value. Clients will continue building capabilities internally.

Equity sounds exciting too, but a lot of the time it’s simply unpaid work wearing a cute Burberry trench coat. Most startups fail, positions remain illiquid for years. And a ten person company doing client work isn’t automatically a great venture investor.

But notice which parts of the model those risks attack.

The execution layers, the productized layers, the things AI will inevitably commoditize. The parts that survive are the parts that were never about execution in the first place.

The ideas, the judgment, the thinking, the relationships, the things you own.

AI is going to become table stakes. The company that lasts will make money from what AI can’t touch and use AI to deliver it at a cost that used to be impossible. The more execution gets automated, the more valuable modern judgment becomes.

So, can a ten-person agency make $100 million?

Yes and no.

No, not today was the consensus. Not inside the current agency model. Not by selling retainers, charging for hours or squeezing another point of utilization out of the same spreadsheet the industry has been drilling for decades.

But as a direction? Absolutely.

The first ten person company to clear $100 million won’t get there by building a better version of the agency model. It will get there through diversified revenue streams and small groups of highly capable people who can think, build, advise and create. And yes, being AI native.

In other words, it won’t win by becoming a more efficient agency.

It will win by becoming something else entirely.

Which is probably why the first ten person company to make $100 million from advertising won’t call itself an agency at all.

Booooo! BOOOOOOOooooo!!!

We have all seen the memes, read the stories, heard the chatter. Over the last few weeks three graduation speeches totally bombed in almost the exact same way.

The Stage

May 8, 2026 at a UCF commencement. Gloria Caulfield is three minutes into her speech to the College of Arts and Humanities when she tells the graduates that the rise of artificial intelligence is the next industrial revolution. The room boos and someone yells “AI sucks.” She’s genuinely like WTF.

The next day, Scott Borchetta walked onto the stage at Middle Tennessee State, where there is literally a college of media named after him, and told graduates that AI is rewriting production as they sit there listening. He got epically booed. His response was to smile and say “Deal with it. Like I said, it’s a tool.”

By the time Eric Schmidt walked onstage at Arizona a week later, the grads were already warming up their vocal cords. Schmidt told graduates AI would “touch every profession, every classroom, every hospital, every laboratory, every person and every relationship you have.” The crowd was so loud that at one point he had to plead with them to let him finish a sentence.

Three stages. The same guttural reaction.

The first lesson is rather simple…know your audience.

These were arts, humanities, media, creative and production graduates entering the workforce for the first time after watching their fields get vaporized on LinkedIn over the last two years. Challenger Gray has tracked over 104,000 US job cuts where companies explicitly cited AI as the reason, across 2025 and the first four months of 2026. The real number is almost certainly higher — that figure only counts the layoffs where the company admitted AI was the cause.

The grads were not booing the AI topic. They were booing being told their fear was a passing weather pattern by rich executives who will not be living inside that storm.

Compare it to the week before where Jensen Huang gave the commencement at Carnegie Mellon. He talked about AI - lots and lots of AI. Zero boos.

Carnegie Mellon kids are studying computer science and engineering, they see AI as something they are going to be part of and eventually lead. The UCF grads, the Arizona class, the MTSU students see AI as something that is going to be done to them.

One room heard opportunity. The other heard replacement.

The Office

Now look around your own company.

A CMO at a brand you have for sure heard of told me last month she had not slept properly in four months. The night before our call she was up at 4am rewriting a board deck, trying to reverse engineer her team’s AI tooling decisions into something that resembled a coherent strategy.

She does not really believe it is one. Increasingly, neither do most people around her.

There is very little room inside organizations right now to openly admit uncertainty, so the fear gets redirected instead. It surfaces as performative confidence in all-hands meetings, leaders standing onstage declaring AI the greatest opportunity of our generation while privately wondering whether any of next year’s assumptions still hold. Or it gets buried inside another hastily assembled AI task force that mostly exists to prove the company is “doing AI.”

News flash, the 22 year old grad booing and the 47 year old CMO rewriting the board deck at 4am are literally the same person.

Both have watched parts of the career ladder suddenly become unstable. Both are being told this moment is exciting by people who already safely made it to the other side. Both are using the technology while being judged on how they use it. Both suspect the rules of the game have changed before leadership has figured out how to explain the new ones.

The boo outside is louder. The boo inside is way more expensive.

This is the oldest move in tech. Hand a transformative technology to the people it will change the most, sell it to them in a language only the people who built it speak, and act surprised when they push back. The internet did this so did the iPhone. AI is just doing it louder, and faster, and at a scale we haven’t seen before.

The Memo

Two weeks ago, Cloudflare CEO Matthew Prince published a Wall Street Journal op-ed explaining why he had just laid off twenty percent of his company despite record revenue. He divided his workforce into three categories borrowed from Peter Drucker. Builders. Sellers. Measurers.

“The vast majority of those we laid off last week were measurers,” he wrote, defining measurers as middle management, finance, legal, internal auditing, and revenue recognition.

That word did so much more damage than the layoff itself.

He just reduced eleven hundred human beings to a label, he invented a category of person whose job is to measure. He told the world the measurers are obsolete and he gave every employee at every other tech company a new way to look at their colleagues.

Are you a builder, a seller, or a measurer? If you had to think about it, you are a measurer. Welcome to the shit list.

This is a smart CEO and the op-ed is well put together, it makes objective sense. The job cuts were the honest part, the framing was the epic failure. Nobody trusts a leader who turns them into a category.

This is no longer just a technology transition. It is a communication crisis.

The Messaging

Gavin Baker (CIO at Atreides Management) told a story on a podcast recently that made me think little a little deeper.

A friend of his has a daughter born with a rare genetic condition. He refused to accept the prognosis, spent months researching with LLMs, and found a safe drug already on the market that nobody had thought to apply. It worked.

There are thousands of these stories - but you have to really go looking.

You do not hear any of those stories on the commencement stage. You do not hear them from the LLM CEOs doing the catastrophic risk media tour, you also do not hear them in any media outlet that matters.

Shyam Sankar at Palantir nailed it - he said the industry is breathlessly asking model makers what they think. Maybe we should be asking the people who actually use the models.

Those are the stories leaders keep failing to see.

The job losses are real, no denying it. And the fear is rational. None of these stories cancel out the huge amount of people that have already lost jobs to AI in the last 18 months, and the next 18 will be worse before they are better.

People will be ragging on it, but humans also tolerate enormous change far better when they feel respected inside it. What breaks trust is not uncertainty itself, it is being asked to perform enthusiasm while your fear is treated like irrationality. That just straight up pisses people off.

The Speech

So now my final couple of questions.

If you had been the one walking onto the UCF stage that morning, what would you have said?

How would you name the fear in the room without dismissing it?

How would you explain the opportunity AI is creating without making people feel disposable in the process?

Take two minutes and write it out. Now the harder part, take that speech you just wrote and stress test it:

Would you give the same speech to your marketing team on Tuesday morning?

Would they believe you?

Would you still say it if the press quoted it back to you word for word?

This is the most exciting moment and the most disruptive moment of employees’ lives, and both things are true at the same time. Most leaders are flunking it because they pick one and ignore the other.

The AI train is moving. We do not get to vote on whether it happens but we do get to vote on whether the people in the actual rooms get a leader brave enough to give them the speech the grads never got.

Be the one who gives it.

Most leaders I know spent twenty plus years climbing to the top. The reward was getting further from the work. That was the deal. Once you reached the corner office with the view, you finally made it. This was when you got to tell people what to do, approve things and come into the big meetings to create some razzle dazzle.

For most of modern corporate history, leadership prestige was tied to abstraction. The higher you climbed, the further you moved from the actual doing. Junior teams made the thing, mid level teams made the deck about the thing and senior leadership approved the deck. By the time you reached the corner office, proximity to execution had almost become a status risk. Demeaning, even.

That system made sense for the era it was built in.

The post war corporation was fundamentally a coordination machine: huge organizations producing standardized output across vast geographies, in a world where information moved slowly and decision cycles ran in quarters, not minutes. Leadership rewarded people who could create order, reduce ambiguity and maintain predictability at scale.

William Whyte called the resulting archetype The Organization Man in 1956. Sloan Wilson (the famous author) dressed him in the gray flannel suit a year earlier.

The person that system rewarded had a specific temperament : controlled emotion, controlled ambiguity, controlled information.

Treated being visibly mid thought, or taking too long to answer a question, as a potential failure of executive presence. Executives perfected the blank face card. You proved you could be trusted with capital because you projected consistency at all times.

They are called “The Ruler’”

Brand strategists will already know that name, because Mark and Pearson built the archetype system for brands twenty five years ago. The Ruler became the dominant leadership archetype of the post war corporate era for a reason.

The system needed steadiness, and it bought steadiness by demanding emotional uniformity at the top. Predictability cascaded. And so did the suppression.

The Ruler has been losing ground for the last couple of decades. Post heroic leadership, founder mode, design led CEOs (Hi Apple and Airbnb). Each has been chipping away at the archetype since the early 2000s.

My hot tip is that AI finishes The Ruler off.

Not because every executive suddenly needs to become an individual contributor, but because intelligence and execution are rapidly becoming more accessible while judgment, taste and creative conviction are becoming far more valuable.

You can already see the organizational version of this shift happening in real time.

Coinbase leadership has spoken publicly about reducing layers and moving toward more player-coach dynamics. Nvidia runs with sixty direct reports under Jensen Huang and no one-on-ones, minimizing management layers in favor of direct access to information and faster decision making. And Brian Chesky still sits deeply inside design reviews at a company worth a bazillion dollars because, as Jony Ive once told him, “you don’t manage people, you manage people through the work.”

Most of the conversation around AI and leadership has been focused on org charts. Flatter teams, fewer layers, agentic workflows. The shrinking executional middle.

But the more interesting shift may just be psychological.

Deloitte’s human capabilities research identified curiosity as one of the defining traits of high-performing teams moving forward. McKinsey has increasingly emphasized “inside-out leadership,” arguing that self awareness and emotional adaptability are becoming critical executive differentiators. Even the broader business conversation is starting to converge around the same idea: as intelligence becomes more abundant, human judgment becomes more scarce.

We now have machines that can generate polished summaries, structured thinking, optimized drafts, competent presentations and synthetic confidence at near infinite scale. Agents are becoming the norm. Robotics is going to be a fast follow.

We spent forty years asking humans to behave more like machines. Now we have the machines, and the thing they cant reliably do is be human.

It may not be perfect, and it may not happen all at once. But some of the traits that make people compelling humans — curiosity, divergent thinking, agility, conviction, humor, hustle, creative energy — will start to become economically valuable at the top of organizations.

In Pearson’s system this new age leader would sit at the intersection of the “Magician” (vision, pattern recognition, transformation) and the “Creator” (hands on craft, making things of enduring value). Either one alone is incomplete for this new era. The Magician without craft is a visionary with a great deck and the Creator without vision is a builder of pretty things.

The shift is towards leaders who can do both, at scale.

I’m calling them “The Shaper.”

The Shaper stays cognitively alive inside the work itself while still shaping what comes next.. They do not climb the ladder in order to escape the making, they climb it to influence it at greater altitude.

The Ruler treated the title as the destination, where as The Shaper treats it as infrastructure.

The Ruler optimized for polished certainty. The Shaper is comfortable thinking publicly, iterating and changing their mind if its the right thing to do.

Not because vulnerability is now fashionable, but because in an AI environment, visible cognition will become trust building. People increasingly trust leaders who feel genuinely engaged with real world over leaders performing rehearsed omniscience.

That’s why so many modern leaders who feel culturally resonant right now land differently than the industrial old archetype. Less executive theater, less institutional polish, less obsession with appearing frictionless at all times.

That does not mean charisma replaces competence. Or that every executive suddenly needs to become a creator economy personality. Most organizations still need structure, coordination, operational excellence and management discipline. They always will.

But the center of gravity for the new type of leaders, is moving.

The character stack of The Shaper, the new AI era leader:

1. Loose grip on status. The Ruler held the title close. The Shaper uses it as a tool - they are curious about the actual work, not the rung.

2. Vision x Work. The Ruler set direction from a slide. The Shaper sees where the work needs to go and is in there getting it done.

3. Conviction, held in public. The Ruler waited for the multiple research decks, reviews and workshops to decide. The Shaper calls it fast, holds the line, and adjusts when wrong.

4. Productive restlessness. The Ruler oversaw the system. The Shaper keeps poking and prodding at it - reshaping.

5. Visibly alive. The Ruler performed gravitas. The Shaper projects intellectual energy, curiosity and genuine engagement with the work. Reads the room & lets the room read them.

6. Plays well with all types. The Ruler managed a handful of humans. The Shaper directs a diverse and larger team of humans and agents together, knowing which problems belong to which.

The industrial era decoupled leaders from makers. This AI era is collapsing them back together. And honestly, I am here for it.

I listened to Evan Spiegel on Lenny’s podcast last week and one idea stuck. He said “Humanity is far more important than the technological developments, largely because humanity dictates how technology is adopted.” He went further, saying he’s expecting “a huge amount of societal pushback on a lot of the changes that are coming with AI.”

So the question I wanted to unpack this week.

What do humans and AI actually look like living together inside a brand experience? And can we build it in a way customers don’t just tolerate, but actually want, and maybe even love?

The situation.

There’s a real tension sitting underneath all of this innovation, and it’s easy to ignore if you’re deep in the AI positive bubble.

A March NBC News poll put AI’s net favorability at minus 20 among US registered voters, 26% positive and 46% negative. Billion Dollar Boy ran the same study in 2023 and 2025 — preference for AI generated creator content versus human creator content collapsed from 60% to 26% in two years. Other studies suggest a meaningful share of people actively prefer brands that don’t use generative AI in customer facing content.

Merriam-Webster’s word of the year for 2025 was “slop.” Ha!

This isn’t some soft sentiment dip, this is a real concern and a very clear signal. And when you pay attention you see it is not evenly distributed, it’s class shaped.

Edelman’s 2026 Trust Barometer found 54% of low income and 44% of middle income consumers globally believe generative AI will leave them behind, and not benefit them. The same NBC poll measured the divide directly inside the US. Upper class voters were the only economic group with positive feelings about AI. The people most enthusiastic about AI are largely the people whose work and wealth depend on it. The people most resistant are largely the people who feel AI is being done to them, not with them.

David Friedberg, on the All-In pod a couple of weeks ago, put it really well. He called the AI data center “the modern temple of the wealthy.” A concrete symbol of elite progress in an age when most Americans feel they have nothing to show for the last twenty years of growth.

You can see (and feel) the same structural argument in the rise of populist movements on both ends of the political spectrum, in city-by-city pushback against data center construction, in the new appetite for billionaire taxes, in the increasing shift of “made by humans” from a niche signal to a mass market positioning lever. These are just a few.

Now flip that, the other side of the divide is moving in the opposite direction at full speed. Solo founded startups jumped from 24% of new US companies in 2019 to over 36% by mid 2025. Vibe coding is a verb now. Anthropic’s CEO is openly betting on the first one person billion dollar company landing in 2026. The AI positive class isn’t just optimistic, they are building fast and with fewer people every quarter.

One side of the divide is being told AI is the future, but they don’t want it. The other side is using AI to write that future, and they wholeheartedly believe it’s the only answer.

The ground under your brand’s AI strategy isn’t stable. Its not a gentle ripple either, it’s a 8.0 earthquake and it’s shifting fast — culturally, economically, emotionally. If you don’t see that clearly, you’re going to build something that works technically, but fails the moment it hits an actual living breathing human.

Tech moves fast. Most people don’t.

Spiegel’s point is simple, the more tech focused crowd thinks people will pick up AI as fast as the AI itself is getting built. News flash. They probs won’t.

AI moves so damn fast - models, chips, money. Keep adding more and it will continue to get smarter and faster but the people side doesn’t work like that. Whether someone trusts a piece of technology, whether they feel it makes their life better or just makes them feel small, will take more time. Building a bigger computer doesn’t make your customer feel better about being served by a bot.

Jensen Huang, the CEO of Nvidia, calls AI a five-layer cake. Energy, chips, infrastructure, models and at the top, applications. He’s said the application layer is “the layer that has to succeed most,” and the one “whose users will benefit from this industrial revolution most.” It’s the layer that brands have to win, its the actual products real people use. (I wrote a deeper piece about that AI cake last week).

The top layer, the application, is the only place where AI meets an actual human being in the wild - the customer, employee & audience. It’s the place where the speed of technology runs into the speed of being a human.

Mass trust will be slower to adopt because AI is in a lot of ways, very personal. Every time AI shows up in someone’s life, it’s subtly asking them a question about themselves, big or small. The customer service team being replaced by a chatbot is being told what they’re worth. The audience watching an AI-made ad may feel the brand doesn’t think human creativity is worth paying for anymore. AI doesn’t just turn up and do a job better, it can change how the person on the other end feels about their own value.

People might not articulate it like that. But they absolutely feel it.

Tech leaders are treating AI like every other piece of new tech before it. Build it, ship it, wait for people to come round because this is freakin’ epic. And honestly, that worked for things like the internet, it worked for smartphones, it even worked for cloud software. But it’s not as clean cut with AI, because AI isn’t just a tool people pick up that helps make their currently life instantly better. It’s a tool that will change the future and with that it can also change peoples identity. From how they think about themselves at work, in their craft, to at home.

And no matter how fast we move on the tech and deployment, on the other four layers of the AI cake, if we can’t get the application layer right, none of the rest of those layers will really matter.

Marketing leaders, its your time to shine.

Two tracks.

Most companies are treating AI adoption as a single decision. News flash, it’s not.

Track one: protect the human moments.

This is human work that should never be replaced. The moments people came to you for, its the reason they pay your premium. Its why they keep coming back time and time again. AI’s job here is not to replace humans, it’s to make the human at the core better.

Take a look at what sports has taught us.

The NFL continues to dominate U.S. viewership. The 2025 NFL season averaged 18.7 million viewers per game. Live sports make up the overwhelming majority of the most-watched broadcasts every year. Women’s sports are growing fast (up 25% year-on-year), with rising revenues, valuations, and expansion fees across leagues.

In a world flooded with synthetic content (check out the ‘dead internet’ theory), the most valuable thing you can watch is still a real person doing something unpredictable, in real time, with real bodies, real stakes, real consequences and with real emotion.

That tells you something about where humanity belongs in your brand experience. People don’t want less humanity, they want more of it — in the moments that matter.

No one wants a chatbot calling the Super Bowl. And they don’t want one handling the moments in their lives that feel just as real.

The brands getting track one right aren’t removing humans from the experience. They’re using AI to make those humans more visible, more capable, and more effective.

AI as amplifier, not substitute.

Track two: build the AI-native value.

Then there is the work people don’t want to do, don’t enjoy doing, and would happily hand over to a machine. AI’s job here is to be the product, super useful and sometimes invisible enough that the customer forgets it’s even there. This is the moment where speed, convenience, and utility matter more than human interaction.

Here, AI should take over, not as a feature but as the product itself.

The best analogy is electricity.

People in the 1880s didn’t trust electricity. It was invisible, dangerous, seen as something for industry and the wealthy. What changed adoption wasn’t a better explanation of what electricity was, or even what it could do etc etc. The proof was in the pudding, the products that made life obviously better — light bulbs, street lamps, the things people could feel immediately.

AI is in its electricity moment.

Most marketing AI today is the equivalent of asking customers to be excited about generators. Fun times. Nobody is going to adopt “AI” because someone told them to, especially if those people are the ones half of society don’t trust and are actively opposed to.

People will mass adopt what AI makes possible. But only the parts they actually want.

Spotify doesn’t ask you to feel excited about its recommendation engine. It just plays you the song you didn’t know you even wanted. The best AI-native consumer experiences feel less like AI and more like ahhhh, finally.

So that’s the second track, build the streetlamp version of AI for your category or brand. The product or service that uses AI to make a customer’s life materially better in a way they couldn’t get before. Value people can feel instantly, without needing to be convinced.

Permission and proof.

OK stripping it all back and wrapping this piece up.

Track one is permission. The customer needs to know you didn’t hand the human moments to a machine.

Track two is proof. The customer needs to feel AI making their life better — not just your margins better.

Because if you get this wrong, AI can make your brand feel cheaper or just plain untrustworthy. But if you get it right, it can make your brand feel more human than it ever did before. That’s the paradox and that’s the marketing leaders job.

That’s the bridge from tolerated to wanted. From wanted to loved.

Human led. AI powered.

Every layer of the AI cake has to succeed or the whole thing falls over.

Now pull a number against that, MIT’s State of AI in Business 2025 report found 95% of enterprise generative AI pilots delivered minimal measurable P&L impact across $30 to $40 billion in enterprise AI spend, with only 5% creating meaningful value.

The 5% or so tend to have something in common McKinsey’s 2025 State of AI survey found they redesigned the workflow around the AI, they didn’t bolt AI onto the workflow they already had.

Most of the failure MIT describes is happening at the top of Jensen’s cake, at the application layer, not the four layers below it. The lower layers are being actively built and fought over - energy, chips, infrastructure, models - and while none of them are by any means trivial, they are progressing fasssst.

Yet enterprises still can’t turn it into returns.

MIT frames it as both a tooling problem and a learning problem. The AI doesn’t retain context, and the organization doesn’t teach it, both are true. But the organizational side is the one marketing leaders can actually move on this quarter.

Ethan Mollick made the same point last week on LinkedIn. In every big company he visits, a handful of senior people “absolutely get AI.” The problem is scaling that knowledge “throughout an organization where many people are further behind.” The gap between the few who get it and the rest of the company is the clear tension.

Most senior marketers are having the AI conversation one layer down from where the win is. They’re debating which model to pick, which vendor to sign, which pilot to fund. The decision that actually matters is how human judgment gets wired into the application layer above all of it.

Here’s the “cake” - vanilla with chocolate buttercream frosting to be precise .

Layer 1: Energy

Jensen said it kinda plainly, the lowest layer is energy. The IEA projects global data center electricity demand will more than double by 2030, driven almost entirely by AI. Permitted, powered land is the binding constraint on the whole stack, it sets the cost floor of every AI tool you buy. When your martech vendor raises prices twice a year and cites “infrastructure costs,” they are passing through energy economics.

What this means in a sentence: No power = no AI.

Layer 2: Chips

NVIDIA lives here, and according to Jensen, so does most of the near future. Every AI application your team ships this year was trained and is served on hardware from roughly four companies. That concentration is why procurement conversations with every martech vendor now include “compute tied pricing” and “AI credit packs.” A new dependency that doesn’t show up on your vendor evaluation spreadsheet.

What this means in a sentence: Compute scarcity shows up in the price of every AI tool you own.

Layer 3: Infrastructure

The hyperscaler layer - AWS, Azure, Google Cloud. If chips are the raw material, then infrastructure is the landlord. Three companies host most of the frontier model labs that power most of martech. OpenAI runs primarily on Microsoft Azure, with Oracle now in the mix. Anthropic runs on AWS and Google. Capability flows to the infrastructure layer first and reaches your stack years later, shaped and priced by companies with no stake in whether your brand looks good using it.

What this means in a sentence: The cloud layer sets the rent and most companies are just tenants.

Layer 4: Models

The layer everyone fights about - GPT, Claude, Gemini, Grok, Llama - new versions every few weeks now, not every quarter. The frontier stays uneven, Claude wins on some things, GPT on others, the rankings flip with every release.

Your peers are spending senior meeting time picking the ‘perfect’ one, they shouldn’t be. Model choice isn’t one decision, it’s a portfolio. A reasoning model for the hard stuff, a fast cheap one for the easy stuff, an open source model for the work you don’t want sent outside your walls (harder at enterprise scale, but worth knowing about). Anything fine tuned to a single model in April gets rewritten by October.

What this means in a sentence: Don’t marry a model. Date a few.

Layer 5: Applications

This is where your brand lives - where AI meets a customer, a product, a campaign, a service. Jensen called this “the layer that diffuses into society”. The winners of the AI era will be the companies that turn that infrastructure into something a customer actually touches. For a marketing leader, that’s the entire game, the model you pick doesn’t build a brand or the cloud you’re on doesn’t build a brand - what you do with them does.

Layer 5 in a sentence: This is where AI reaches and impacts a customer. Or doesn’t.

Agents book meetings, draft briefs, file tickets - the question isn’t whether humans stay in the loop, it’s which loop. Which decisions still need a person at the top, and which don’t. Most marketing organizations haven't made that call yet, and from what I'm seeing, that's a big gap.

Strategy, creative judgment, taste, and the call on which AI output ships and which gets killed all stay on the human side. AI has throughput, it doesn’t have a stake in whether your brand wins. The moment a marketing organization reverses the polarity and lets AI drive direction, the application layer collapses into the 95%.

Back to the cake. And now with a cherry on top.

Jensen's point was that every layer has to work, but the applications layer matters the most, it has to turn AI into something people actually use. Energy, you can’t control. Chips, you can’t control. Infrastructure, you can’t control. Models, you can kinda influence.

The application layer is the one you can own, it is everything your customers touch, everything your brand stands for, taste, every judgment call about what’s worth making and what isn’t, lives there. For a marketing leader, it’s the only one that’s yours to truly succeed at, it’s the layer that will separate the winners from the rest.

One easy thing to do this week: Pick the workflow in your team where AI is producing the most output right now. Ask who sets the direction on that workflow, if the answer is “the prompt,” you probably need to take a second look.

Ask someone who works in professional services what they do, most of the time they won’t tell you about the work. They’ll give you a title - SVP of This, Senior Director of That, Head of Something That Used to Mean Something. Then they wait for you to place them on an invisible ladder and congratulate them on their success.

That’s not a personal quirk, it’s a century of conditioning and the container is starting to seriously crack.

Marketing is the function where we may see it clearest, and for a specific reason. Marketing has always run on a mix of in-house teams, agencies, freelancers, and specialists rotating in and out, it was the function that was already half outside the traditional employment container even before AI showed up.

Act one: what’s actually ending

Shopify made the new reality explicit earlier this year, CEO Tobi Lütke posted a memo on X himself because it was about to leak: “Before asking for more headcount and resources, teams must demonstrate why they cannot get what they want done using AI.”

2.6 million views later, a standing ovation from investors, pretty much zero backlash. Shopify’s own headcount dropped from around 11,600 to 8,100 across three years while revenue jumped 26% in 2024 alone, yet nobody called it a layoff.

Every Fortune 500 CEO is now running a version of the same playbook and the ones who say it out loud (hi, Duolingo) learn that framing matters. Lütke framed his memo as ambition, a higher bar before new hiring. Von Ahn CEO of Duolingo framed his as a replacement - same underlying shift, opposite receptions.

The bloated marketing & agency org, with its layers of managers and endless decks and whole sub functions dedicated to moving information around, doesn’t hold up under AI pressure. Not because the work matters less, but because a lot of what filled those org charts wasn’t the work itself, it was coordination around the work, and AI collapses that. Reporting, first drafts, adaptation, internal updates, all the in between steps that used to take time and people and process.

The question shifts from “how many people do we need” to “who actually knows what they’re doing.”

And as execution keeps getting cheaper, judgment could get more expensive. That’s the move underneath everything else, less value in doing the work mechanically but more value in deciding what work is worth doing in the first place, how it should show up - and then ultimately making it happen.

By the time the Shopify memo becomes standard operating procedure (sooner rather than later), the career logic that held marketing together for decades stops working. Average CEO tenure across 1,800+ public companies fell to 6.8 years in the first half of 2025, the lowest in at least a decade. Entire Fortune 500 categories are being restructured faster than Taylor Swift can drop a new album.

The standard marketing arc, 25 years at a big corporation with golden handcuffs pulling you through the vests and retirement on the other side of the pension, relied on the company still being there at the end. That assumption is walking a very wobbly tight rope.

The container isn’t cracking for one reason, it’s cracking because every load bearing wall is under pressure all at the same time.

Act two: what’s replacing it

The flipside of fewer full time jobs is the rise of a different kind of marketing entity - solo operators, fractional leaders, small micro companies of ten or fewer people who can do what used to require entire departments or agencies. Those that understand this new world order, human led & AI powered.

And the economics favor them - running your own business means writing off your home office, your stack, and your travel as the cost of doing business. That tastes a lot sweeter than a W-2 paycheck that’s already been taxed to the bone by the time it hits your account. It also means saying yes to the projects that interest you and no to the ones that don’t. The corporate job was the stable choice for a long time, for senior people with real reputation, the operator path may now be the more stable one.

Upwork’s 2025 Future Workforce Index puts more than one in four US knowledge workers in independent or freelance work, generating $1.5 trillion in earnings last year. This isn’t fringe behavior, it’s the working model that may one day become the default.

The commercials shift with it, retainers and hourly billing start to feel a bit off when one person with the right setup can do the work of ten. What’s growing instead is outcome based pricing, productized services, subscription access to senior judgment, fees tied to results. You’re not selling time anymore, you’re selling the answer.

And the work itself changes shape with less servicing, more think × solve × build. A company with a growth problem doesn’t need more weekly status calls. It needs someone to solve why growth stalled, think through the go-to-market reset, and build the measurement system that shows impact fast, in six weeks, with a fixed fee and a custom-built agent delivered alongside the work.

These marketing operators also don’t have to stay solo for long, they can assemble into a loose ‘coalition’ model. A mix of people needed for the brief comes together, does the work, then disperses. Bring in a strategist for this one, a growth operator for the next, a creative lead when the problem demands it. Different micro-companies or different individuals, coming together for a defined piece of work, then disbanding once it’s done. Not a permanent org chart or payroll but a new system that can reconfigure itself based on the problem in front of it.

That’s a similar-ish model holdcos have been trying to build at the top end for years. For enterprise clients with huge budgets, they can make a version of it work, but for mid-tier and founder led brands, the economics just don’t match. The overhead is too high and the speed is too slow to compete with a small senior (and AI native) led coalition.

And to be clear - traditional marketing roles aren’t going to vanish from corporate America, but as the shape changes significantly, there just will be fewer of them. The skills that made someone a great marketing leader in 2018 aren’t necessarily the same skills that will make someone great in 2028.

Act three: what it means, and what nobody really knows

The most uncomfortable shift is at the entry level - junior roles, associate programs, brand management tracks were where people learned the craft. Those roles are the easiest to compress or remove. Some people are building a replacement path in public, using AI as leverage, but plenty won’t.

Before the industrial era, people learned through apprenticeship but the corporate economy replaced apprenticeship with junior jobs mostly due to cost. Now junior jobs are vanishing before we’ve built anything real to replace them.

But I’m not going to pretend the AI-kills-the-middle case is fully settled, there’s a counter argument worth mentioning - MIT’s David Autor argues AI could actually rebuild middle class work by extending expert level decision making to more people. Brynjolfsson and Li’s 5,000 agent study found AI boosts novice productivity by 34% and barely moves top performers. Inside defined tasks, AI lifts the bottom more than the top, I totally buy that finding.

What I’m betting on is that the real re weighting isn’t happening inside tasks though - it’s happening at the level of who decides which tasks matter. Execution gets cheaper for everyone and judgment gets more valuable for a smaller group, that’s the gap I am seeing start to widen.

Income volatility is real, corporate paychecks were boring on purpose. Fractional income comes in pulses and striking out on your own has its own set of risks. The people doing well have the skill, reputation, and financial buffer to ride the variability - but plenty of people who don’t have these still try and give it a go.

Inequality almost certainly widens before it narrows, the top end of any marketing discipline captures more value faster, while the middle thins out. Marketing used to offer a reliable middle class path but that path is getting narrower, and I should be honest that I’m writing this piece primarily for the people on the winning side of that shift, it will be painful for many others. The harder question is what we owe the people who aren’t going to win this new world order, and I don’t have a clean answer for that yet.

Belonging is the piece people talk about least, marketing and agency teams used to be where your tribe lived - offsites, war rooms, big wins, big losses etc. The epic and wild agency culture you only understand if you survived it. The shoot everyone still talks about because something caught fire, metaphorically or otherwise. That doesn’t automatically exist when teams are smaller, more fluid, and partially replaced by tools. People are rebuilding it in networks, communities, and peer groups. It works but I’m not sure if it is quite the same.

Act four: Clear Implications

If you’re running a large marketing function, the honest question is whether your org was designed for a world where AI exists, or whether it just grew over time. Most grew, most are now culling, most are still the 2015 structure with AI bolted on. The teams pulling ahead look less like a pyramid and more like a connected ecosystem - smaller senior cores (who can think and do), rotating specialists, integrated operating system, and AI doing the coordination layer etc.

If you’re a senior marketer wanting to take a leap for yourself, the shift is in your favor, as long as you’re actually good (brutal real talk). Stop thinking like an employee and start thinking like a portfolio. Your reputation, tangible output and your network pay you forward from here, your title doesn’t.

If you’re operating a mid sized or founder led brand, this is more opportunity than threat. You don’t have the legacy structure to unwind and rebuild, you can build this way from the start and help future proof yourself. The opportunity to win here is massive.

Act five: Final thoughts

The ‘work’ itself isn’t going anywhere, if anything, there’s more of it coming. As AI makes execution cheaper, the number of things it becomes economically viable to do expands dramatically. Markets that were too small to serve get served, problems that were too expensive to solve get solved, founders who couldn’t afford a marketing team can now run one with two people and a stack of agents.

The work that’s disappearing is the bottom half of the task list - the work that’s appearing is everything that requires taste, judgment, synthesis, relationship, original thought and yes, AI understanding.

And then there will be a whole category of work we can’t even name yet. Entire new disciplines, needs that don’t currently exist, problems nobody’s framed because the tools to see them haven’t arrived. Some of the best careers of the next decade will be things that weren’t on anyone’s org chart in 2026.

The job, as we built it, is what will eventually look temporary. We didn’t invent work in the 20th century, we invented payroll architecture, useful for its time, but built for a world that’s already moving on. What comes next is hard to accurately predict, but I do think it will look less like the job as we know it today and more like a portfolio, less like a single career path and more like a practice, work tied to what you can do and not what title you have. The thing a job used to mean, the stable five day a week lifetime contract, will be one option among many, and may not be the default.

Marketing may just be one of the functions that’s going to feel it the earliest, and has to answer the harder questions for everyone else.

Human Led. AI Powered

Sources:

Shopify annual reports and SEC filings, 2021-2024.

Russell Reynolds Associates, Global CEO Turnover Index, H1 2025.

Upwork Future Workforce Index 2025; MBO Partners State of Independence 2025 (15th annual report).

Stepstone Group analysis of 4+ million job ads, Q1 2025; Randstad analysis of 126 million job postings worldwide, 2024-2025.

New York Federal Reserve labor market data, September 2025.

SignalFire State of Talent Report 2025.

World Economic Forum, Future of Jobs Report 2025 (survey of 1,000+ companies representing 14M workers across 55 economies).

David Autor research: “Applying AI to Rebuild Middle Class Jobs,” NBER Working Paper No. 32140, February 2024.

Brynjolfsson, Li & Raymond study: “Generative AI at Work,” Quarterly Journal of Economics, May 2025 (5,172 customer service agents studied).

McKinsey Global Institute, “A New Future of Work: The Race to Deploy AI and Raise Skills in Europe and Beyond,” May 2024.

That piece was about what changed with the customer journey. This one is about what changed for your website.

Most marketing teams still treat the website like a destination - rank, click, visit, browse & convert. That model is done. Like, done done.

McKinsey’s report on AI search put numbers on it that IMO should have caused immediate action from marketers. About 50% of Google searches already serve an AI summary, by 2028, that number is expected to pass 75%. Roughly $750 billion in US consumer spending will flow through AI powered search.

That is not a channel tweak, that is a structural change in how discovery works.

Another stat that will make you face plant, Forrester’s 2025 Buyers’ Journey Survey found that generative AI tools are now the ‘single most cited source people use when researching a purchase’. Not vendor websites, not product experts, not analyst reports, but AI.

Yet 20% of those same people said the AI gave them information that was unreliable or flat out wrong. So the machine is the new first impression, and yet it is getting things wrong.

The website used to be the destination, now it is the source. If AI systems are doing more of the reading, summarizing, comparing, and recommending, your website is no longer the place someone lands - it is the raw material the machine pulls from before they ever get there.

That changes what website content is for.

A lot of brands still treat content like a traffic machine, with generic articles, keyword chasing, a blog full of beige opinions nobody could disagree with. That approach is dead and nobody sent flowers.

AI systems do not reward sludge, they reward clarity, structure and a real point of view. Something clean enough to cite and strong enough to trust. McKinsey found only 16% of brands even track how their content performs in AI powered search. Most companies are flying blind while the ground shifts under them.

Pernod Ricard found out the hard way, they partnered with an agency to audit how LLMs were representing their spirits portfolio. What they found was a mess, one major AI model was telling people that Ballantine’s Scotch was a prestige product ,it is not. It is an affordable, more mass market brand, the positioning was literally backwards. Any price sensitive person asking an AI about affordable Scotch would never see Ballantine’s in the answer. That customer was lost before Pernod Ricard knew they existed. Harvard Business Review published the full story earlier this year. Pernod Ricard now runs regular audits of what they call “share of model,” prompting every major AI platform, cataloging responses, and rewriting website and ad copy to correct the record.

They are not the only ones doing this manually. Ally Financial’s CMO Andrea Brimmer has her team running weekly scrums specifically to monitor how Ally shows up across LLMs. When they find AI systems citing outdated or incorrect information, they build new content and work with comms to correct the original sources being referenced. Brimmer recently described the current moment as “hand to hand combat” to keep brand accuracy intact across probabilistic chatbots.

Two big brands both discovering the same thing, if the machine gets your story wrong, the market gets your story wrong - you don’t just lose the deal, the deal never even existed.

To be clear, content clarity alone will not solve this. Domain authority matters, third party mentions matter, how the broader web talks about you matters - but those are longer plays. What you can control immediately is what your own site gives the machine to work with.

Three things follow from this.

First, the job of a website is shifting from persuasion to evidence. Your homepage still matters, case studies still matter. But AI systems are not persuaded by vibes, they pattern match for usable proof like clear claims, specific expertise, named sectors, defined problems, actual positions. If your site says you help brands grow in a changing world, you have given the machine nada. If your site says you help CMOs redesign marketing operating models for AI led discovery, that is something a system can use.

Second, structure is now part of strategy. Messy websites used to just be annoying, now they are a full on liability. If a page rambles or talks in broad agency waft, the machine has to guess. When it guesses, you get a problem. Your brand shows up as something it is not, or it does not show up at all.

Third, point of view is now infrastructure. Most brands think authority comes from publishing more, it doesn’t. Authority comes from publishing something actually ownable. In an AI search environment, brand is not a soft layer sitting above performance, brand is the memory structure. It is the set of associations the machine finds, repeats, and uses to decide whether your name belongs in the answer at all. Vague positioning, vague visibility - your brand is literally beige.

The website is no longer the destination, it is the source document AI systems read to decide whether your brand is credible enough to mention.

If you want a Friday morning move, paste your top service page into an AI chatbot and ask it - what your brand does, who it serves, and what position you hold that a competitor wouldn't. If the answer is vague or could describe anyone else in your category, that's a (fixable) problem.

The website still matters, just not for the reason most people think.

frankiejeangroup.com

You’re a senior marketer who’s been in the game twenty years. Your team is full of people who throw around words like “agent” and “vibe code” and “OpenClaw” in standups, and you nod at the right moments. You have ChatGPT and you’ve built some decks with it. It’s helped you write some stuff, but you mostly use it as a Google substitute.

You know, at some level, that this AI moment is the most important shift of your career. You also know the gap between what you actually understand and what you’re supposed to understand is getting wider every month.

You don’t want to ask in a big meeting “wait, what’s the difference between a chatbot and an agent,” because the whole room is watching you for signals about whether the company is going to be okay. So you Google it at 11pm, you kinda read someone’s LinkedIn post. But honestly, the amount of AI slop out there is overwhelming, and you don’t really know who to trust or what to read these days. Exhausting.

I’ve talked with so many people who all tell me the same thing. If that sounds like you, read on.

I wrote a super simple, plain talking cheat sheet of some key terms, an explainer on what an agent is and what it means to “build” one, the main platforms people are talking about right now, and eight things you can do this weekend to stop feeling behind.

The basic cheat sheet

Prompt. What you type to the AI. When someone says “good prompting” they mean writing clear instructions with enough context.

System prompt. Background instructions that tell the AI who it is and how to behave, before you even start typing. The rulebook the user never sees.

Chatbot. An AI you have a back-and-forth conversation with. ChatGPT, Claude, Gemini. One turn at a time, you’re driving.

Agent. An AI that takes multiple steps on its own to finish a task. You give it the goal, it figures out the steps, goes and does them, and comes back when it’s done. More details below.

LLM (Large Language Model). The brain underneath ChatGPT, Claude, Gemini. “LLM” and “the model” are used interchangeably.

Hallucination. When the model confidently makes something up. A fake quote, a fake case study, a fake statistic. Happens a lot. The real word for it is “lying with conviction.” Always verify anything that matters.

Context window. How much information the model can hold in its head at once. Short-term memory size.

Token. A chunk of text or code the model reads and writes, roughly four tokens per three words of English. That's how usage gets counted on your bill & they add up fast.

Reasoning model vs chat model. A chat model (GPT-4o, Claude Sonnet, Gemini Flash) is fast and great at language. A reasoning model (OpenAI’s reasoning models, Claude with extended thinking, Gemini’s deep thinking modes) actually thinks before it answers. Slower, pricier, dramatically better at hard problems. Chat model for writing and recaps, reasoning model for strategy and analysis.

Deep research mode. A mode in ChatGPT, Claude, and Perplexity where the AI goes away for 10 to 30 minutes and comes back with a full report. Use it on questions you’d otherwise pay a freelancer for.

Multimodal. The model can handle more than just text. Images, audio, video, PDFs. Upload a screenshot of a slide & ask “what’s wrong with this” and it can actually see it.

Open source vs closed source. Closed source models (GPT, Claude, Gemini) you rent from a company. Open source models (Llama, Qwen, DeepSeek) you can download and run yourself. Matters when client data privacy comes up, which it will.

Frontier lab. The handful of companies building the most advanced models. Anthropic, OpenAI, Google DeepMind, xAI, Meta. “Labs” is the shorthand.

Compute. The raw computing power (mostly Nvidia GPUs) used to train and run AI models. The oil of the AI economy.

RAG (Retrieval-Augmented Generation). The AI looks things up in your documents before answering, instead of relying on what it was trained on. Looking things up, not making them up. NotebookLM is the everyday example.

MCP (Model Context Protocol). An open standard published by Anthropic in late 2024 that lets AI models connect to external tools — your calendar, your CRM, your Notion, your Drive. Most major labs have adopted it.

Python. A programming language, the dominant one for AI. Most AI tools and libraries are written in or controlled through it. When someone says “we’ll just write a quick Python script,” they mean “we’ll automate this in a few lines of code.”

Terminal. The plain text command line on your computer (the little black window with white text). It’s how developers talk to their machine directly, instead of clicking icons.

Vibe coding. Building software by describing what you want in plain English and letting the AI write it. The reason non developers are shipping real products in 2026. Claude Code changed the game.

Eval. Testing whether your AI workflow is actually any good, with real test cases and scoring. Most agencies talk about doing evals. Almost none of them actually do, lol.

AEO (Answer Engine Optimization), sometimes called GEO. The new SEO. As people increasingly ask AI tools questions instead of Googling, brands have to optimize to be the answer the AI gives, not the link the user clicks. You’ll hear this one a lot in 2026.

A quick interlude: wtf is an agent?

A normal AI chat is a conversation. You type, it answers, you type again. One turn at a time, you’re driving. An agent is an AI that takes multiple steps on its own to finish a task you gave it. You set the goal, it figures out the steps, goes and does them, checks its own work, tries again when something fails, and comes back when it’s done. You’re not driving, you’re delegating.

Until about a year ago, agents didn’t really work. They’d get halfway through a task and fall over. In the last twelve months that changed, and that’s the big “aha” moment you’re hearing about non stop.

What it means to “build” an agent

Super simply put, there are three levels.

Level 1. The no-code agent inside a tool you already use. A custom GPT in ChatGPT. A Claude Project. A NotebookLM grounded in your brand material. You’re not coding — you’re loading the context once and reusing it over and over. Time to build: an afternoon. Difficulty: easy peasy. The hard part is being clear about what good output looks like. Already more useful than what most people are doing with AI.

Level 2. The hosted agent with tool access. OpenClaw on a Mac mini, Claude Cowork on your desktop. Now your agent can actually do things in the world like read your email, post to Slack, browse a website, run on a schedule. Still no code. You’re configuring, not programming. Time: a weekend. Difficulty: moderate to hard depending on the specifics.

Level 3. The custom built agent via vibe coding. You describe a real product in plain English to Claude Code, Cursor, or Lovable, and watch it write actual working software. The output is a real codebase that runs as its own product. Time: a weekend for something simple, longer for something serious. Difficulty: moderate to hard depending on the specifics.

You don’t have to climb all three. Most senior people will live happily at Level 1 forever. Level 2 is where it starts to feel magical. Level 3 is where you stop being a marketer who uses AI and start being a marketer who builds with it.

The platforms everyone is talking about right now

OpenClaw. The viral one. An open source personal AI agent platform that lives on your own machine (I run mine on its own Mac mini) and you talk to it through Telegram, WhatsApp, or Slack. It can read your email, manage your calendar, run tasks overnight, and learns new skills as you give it new tools. OpenAI acquired the project in early 2026. The closest thing to a “personal employee” anyone has ever built.

The good: runs on hardware you own (key for data privacy), works 24/7, learns new skills as you give it new tools. When it works, it reallyyyyy works!

The bad: setup is still kinda cobbled together. Nothing’s really standardized, so you’re stitching across models, tools, permissions, and hoping it behaves the same way twice. Most of it is still open source and community maintained. Which sounds great until something breaks at 2am and your “support” is a GitHub thread. Security is the big issue: prompt injection, where a malicious email or webpage tricks the agent into doing something it shouldn't, is currently the biggest unsolved problem in agent security and there's no clean fix, yet. It’s a major reason most enterprise companies won’t sanction agents like this inside their environments.

Claude Cowork. Anthropic’s desktop platform for non developers. You give it tasks on your own computer and it goes and does them across your files, folders, and apps. Closest thing to a built for business people agent from a major lab.

The good: frontier lab model quality, no setup faff, designed for people who don’t want to think about Python. The safer enterprise bet right now. Pretty dope.

The bad: your computer has to be on for it to work, no laptop means no agent. Less always on more “AI assistant for the workday.”

This category is moving fast. There will be a flood of new tools, better security, more enterprise ready versions, all of that is inevitable. The underlying capability is what to understand, not any single product.

Eight things you can immediately do:

1. Run more than one tool, and know when to use which. At minimum, you need both ChatGPT and Claude, both paid. Add Gemini if you live inside Google Workspace. Different models are better at different things, and the only way to know which suits your work is to use them all for a month. Once you’re using two tools daily, learn the difference between chat models and reasoning models, which is the single biggest fix most people need.

2. Use it on personal life first, then bring it to work. Use AI on low stakes life stuff before work, because the cost of looking dumb is zero and the reps build the same. In your first two weeks: plan a family weekend with real constraints, upload a contract before you sign it and have a real conversation with it, prep for a doctor’s visit by uploading your lab results. Then bring the same energy to work — summarize a 60-page research report, turn an interview transcript into clean insight language, pull three strategic platforms out of a messy brief.

3. Verify everything, and learn to prompt with structure. LLMs hallucinate. They’ll invent quotes, case studies, statistics, even whole companies. If it’s going anywhere near a business document, check it. While you’re building the verification habit, learn the five parts of a good prompt: Role, Task, Context, Examples, Format. Anthropic and OpenAI both publish free prompting guides on their docs sites. Twenty minutes of reading, highly recommended.

4. Voice mode, an hour a day, dead time only. Commute, dog walk, dishes. Both ChatGPT and Claude voice are actually good now. Have it explain to you why Jensen Huang and Nvidia data centers are suddenly the most important story in business. Have it interview you about a strategy problem you’re stuck on. The AI infrastructure layer is the single biggest thing most marketers don’t understand about this moment, and a 20 minute voice conversation will be more helpful than skimming a bunch of stuff.

5. NotebookLM properly. Plus deep research weekly. Probably the most underrated product in AI for senior marketers right now. Upload everything for a brand (guidelines, research, past campaigns, transcripts) and it becomes a research partner grounded only in your supplied sources, so it stops hallucinating and starts citing the material you gave it. For a senior person inheriting a brand, it’s the closest thing to a strategist’s brain you can get in an afternoon. Pair it with weekly use of the deep research modes in ChatGPT, Claude, or Perplexity for category scans and cultural tension hunts.

6. Build one Claude Project or custom GPT for a workflow you repeat. Loading context up front and reusing it forever is the jump from chat to system. Easy example: a board narrative project loaded with your last four board decks, your KPIs, your strategic priorities, and your CEO’s communication style, so when you need to draft an update, the voice and framing are already right.

7. Vibe code something small. This weekend. This is the move that changes how you think about AI. Open Claude Code, describe a tiny thing you wish existed, and watch it get built in front of you. Easy example: a private homework tutor for your kid. You load in their school’s curriculum, their recent report card, the specific topics they’re struggling with, and the way they actually learn — visual, verbal, needs things broken into small steps. Then start asking questions to your digital tutor. You are feeling what “build” means now, so you stop nodding politely when someone says “we’ll just spin up an agent for that.”

8. Build a real intake stack. Stop getting overwhelmed by all the noise, focus on a short list of trusted voices in the AI game. Spend some time finding the ones you really like — five sources, properly used, beats fifty sources skimmed. A few examples, Ethan Mollick (One Useful Thing on Substack), Andrej Karpathy (YouTube — start with his free Intro to Large Language Models talk), Hard Fork (Kevin Roose and Casey Newton at the NYT), The AI Daily Brief (Nathaniel Whittemore), and Ben Thompson (Stratechery).

One last thing

You don’t need to become an engineer. But you do need to stop feeling behind, the gap is closeable. Start tonight.

Human led. AI powered.

frankiejeangroup.com

Caveat: this space is moving so fast that I’ll need to update this piece as things evolve.

What’s actually happening is more interesting and harder to fit perfectly onto a slide: the funnel is being compressed. The middle — the part where people research, compare, consider, and decide — is rapidly shrinking. Not because consumers stopped doing those things. Because they’re increasingly handing that work to AI.

A SEMrush study of over a thousand US shoppers found that 51% now use AI during early discovery, 57% use it to narrow their shortlist, and 53% use it at the point of active comparison. BCG’s longitudinal consumer research shows shopping-related GenAI use grew 35% in just nine months during 2025. With 80% of consumers plan to use GenAI to shop in 2026.

Those numbers describe a structural change, not a behavioral trend. The consideration phase — the part marketers spent two decades and enormous budgets trying to own through content, SEO, retargeting, and nurture sequences — is being automated. Not eliminated. Automated. The consumer still goes through consideration. They just don’t do it personally anymore.

That distinction matters. Because it changes what “replacing the funnel” actually means.

There’s no shortage of proposed replacements right now. The smartest ones are worth understanding, because each of them gets something right.

21st Century Brand, writing in Fast Company, argues for replacing the funnel with a flywheel built on what they call “priming and proving.” Priming is the emotional work — brand building that creates memory and trust. Proving is the informational work — credibility that shows up when someone or something checks the facts. In their model, these two forces compound rather than hand off — strong priming makes AI recommendations feel more trusted, and great proving strengthens memories with both humans and machines.

That’s a useful frame. The compounding idea is right. The connection between brand emotion and AI credibility is real and under explored.

BCG takes a different approach. They propose “influence maps” — replacing the sequential funnel with a framework built on four overlapping consumer behaviors: streaming, scrolling, searching, and shopping. Their argument is that these behaviors happen across the entire journey simultaneously, not in stages, and that the degree of influence a brand has at each touchpoint matters more than where in the funnel that touchpoint sits.

That’s also right. It’s the most honest acknowledgment I’ve seen from a major consultancy that the linear model doesn’t reflect how people actually behave. The weakness is that it’s complex to operationalize. Influence maps require real time data infrastructure and AI powered media execution that most marketing organizations don’t have yet.

McKinsey’s position is more foundational. Their research shows that companies blending brand building with performance across the full customer journey achieve 15–20% higher marketing ROI than those focused on lower funnel tactics alone. But their own surveys reveal the gap: 87% of senior marketers say brand building is their most important strategic capability for growth, yet only 58% say their organization is mature enough to actually do it well. WARC

That gap — between knowing brand matters and being able to build it properly — is the real story underneath all of these frameworks.

So where does this leave us?

Here’s what I think the evidence actually supports, honestly, without trying to be clever about it.

The funnel isn’t being replaced by a single new shape. It’s splitting into two distinct zones with a compressed, increasingly automated middle.

Zone one is human. This is where brand gets built. It runs on creativity, cultural relevance, and emotional specificity. It’s the Super Bowl spot that gets discussed at Monday coffee. It's the Amanda and West bombshell turning Summer House into the only thing anyone's talking about this week, and Peacock quietly being the answer every time someone asks an AI where to watch. The brand that shows up inside culture — not adjacent to it — in a way that gives people a real reason to talk. This zone has always existed. What’s changed is what it feeds. Every cultural moment, every piece of creative work that earns genuine human attention, every conversation a brand sparks — that’s now the raw material AI trains on. AI models learn from what people talk about, search for, and share. Every earned mention feeds into the training data that shapes future LLM responses. Brand building isn’t just about being remembered by consumers anymore. It’s about generating enough human signal that machines learn your name too.

Zone two is machine. This is where AI mediates the decision. Gartner projects traditional search volume will drop 25% by 2026 as users shift to AI chatbots and virtual agents. Salesforce reports AI driven traffic to retail sites grew 302% YoY in Q4 2025. When a consumer asks AI for a recommendation, the brands that surface are the ones with enough structured authority, editorial credibility and community conversation across the internet to be confidently cited. Research shows only 45% of brands AI recommends overlap with traditional Google search leaders, which means the rules of visibility have genuinely changed. What makes you findable to an LLM is different from what made you rank on Google. It’s expert citation, authentic community discussion, presence in the places AI actually trains on — Reddit, specialist publications, YouTube, review ecosystems.

The compressed middle is where most marketing spend currently sits — and where the reckoning is coming. The consideration phase hasn’t disappeared. Someone still compares options. Someone still weighs trade offs. But increasingly that someone is a machine, not a human. The content marketing, the SEO, the retargeting, the nurture sequences — all of it was built to influence a person in the act of deciding. If the person has outsourced that act, the infrastructure built to serve it loses value fast.

BCG found that consumers describe AI recommendations as “direct, objective, and personalized, and that AI helps them figure out what they want even when they aren’t initially sure”. That last point is critical. The machine isn’t just answering questions at the point of intent. It’s shaping preference before intent fully forms. That’s territory brand used to own — and now shares with an algorithm.

This has real implications for how marketing budgets should be structured. Most marketers still allocate roughly 70% of their budgets to performance media and only 30% to long term brand building. That ratio was designed for a search dominated world where you could capture intent at the moment it appeared. In a world where the intent capture infrastructure is being disintermediated by AI, that split needs serious reexamination. Not because performance doesn’t matter — it does — but because the inputs to performance are shifting. The brands that AI recommends will be the ones with the strongest human signal. And human signal is built by brand work, not performance work.

None of the existing frameworks are wrong. The flywheel captures the compounding dynamic between brand and credibility. Influence maps capture the non linearity of real behavior. Full funnel integration captures the ROI case for balancing brand and performance. But none of them, on their own, capture the full picture: a world where humans build brand through creativity and culture, machines mediate the decision through training data and authority, and the middle ground between those two is collapsing fast.

The leader who see this clearly will invest more in the kind of creative, culturally embedded brand work that generates real human conversation — because that conversation is now the fuel for machine recommendations. They’ll build a new discipline around AI visibility that most marketing teams don’t have yet. And they’ll have the honesty to admit that a meaningful chunk of what sits in the middle of their current marketing operation is losing relevance, not because it was bad work, but because the consumer handed that job to a machine.

That’s not a framework. It’s a reckoning. And it’s already here.

frankiejeangroup.com

I’ve sat in enough restructure meetings to know how they end. Someone draws a new org chart on a whiteboard. Boxes move. Dotted lines appear. A few titles change. Everyone leaves feeling like something happened. Nothing happened.

The boxes moved. The logic didn’t.

Last week, Sequoia published a piece by Jack Dorsey and Roelof Botha that I think is excellent. Not because it’s about AI, everything is about AI right now, and most of it says nothing. This one says something.

Their argument starts with a history lesson. Organizational hierarchy traces back to the Roman Army — eight soldiers sharing a tent, led by a decanus. Ten of those units make a century. Six centuries make a cohort. Ten cohorts make a legion. The structure scaled because it solved a real problem: humans can only effectively coordinate somewhere between three and eight people. That constraint — span of control — has governed every org chart ever drawn. Roman legions. Prussian war rooms. American railroads. Your marketing department.

Two thousand years. Same constraint. Same solution. Add more people, add more layers.

And then they say the thing people are just starting to quietly realize: what if the hierarchy itself is the bottleneck — not the people inside it?

I spent twenty years inside Fortune 100 marketing structures. I know exactly what hierarchy does to a marketing organization. It doesn’t just slow decisions down. It changes the nature of the decisions that get made. When information has to travel through four layers of people before it reaches the person who can act on it, the information changes shape. It gets smoothed. Softened. Averaged. By the time a customer insight reaches someone with budget authority, it’s a bullet point on a slide in a quarterly deck. The heat is gone.

That’s not a people problem. That’s a plumbing problem.

Dorsey’s argument is that most companies today are using AI as a copilot — making the existing structure work slightly better without actually changing it. I’d go further, in marketing, we’re not even doing that well. We’ve given the content team a writing tool. We’ve plugged something into the media workflow. We’ve sped up reporting by a week. And we’ve called it transformation.

It’s not transformation. It’s air freshener in a burning building.

The real question is this: if you were building your marketing organization from scratch today, knowing what AI can now do, would you build the one you currently have?

Of course you wouldn’t.

You wouldn’t build five layers between the CMO (or agency President) and the person writing the brief. You wouldn’t design a quarterly planning cycle in a world where market conditions change weekly. You wouldn’t create a structure where the people closest to the customer have the least authority to act.

Block is trying something radical. They’re replacing the coordination function of hierarchy with what they call a “world model” — a continuously updated picture of the business that gives every person on the edge the context they need to act without waiting for information to travel up and down a chain of command.

Think about what that means for marketing. Right now, the middle layer of most marketing orgs exists primarily to route information. To aggregate reports, translate strategy into briefs, relay feedback, manage alignment. If a system can do that, and increasingly it can, what’s left?

What’s left is the bit that actually matters. The judgement. The taste. The ability to read a room that data can’t read. The instinct to kill a campaign that tests well but feels wrong. The conviction to back an idea that has no precedent.

That’s the human work. And it’s been buried under layers of coordination and paper shuffling for decades.

Block’s proposed model boils down to three roles: individual contributors who are deep specialists, Directly Responsible Individuals who own specific problems with the authority to pull resources across functions, and Player+Coaches who still build things while developing the people around them. No permanent middle management layer. No standing committees. No alignment meetings about alignment meetings.

I’m not naïve enough to think you can transplant a fintech organizational model into a CPG marketing department next Tuesday. But the direction of travel matters more than the destination right now. And the direction is clear: the organizations that will outperform over the next five years are the ones being deliberately designed for a world where AI handles coordination and humans handle judgement.

McKinsey’s latest numbers show that fewer than 10% of organizations have moved beyond AI pilots into structural redesign. In marketing, I’d bet that number is lower. Not because leaders lack ambition. Because the problem hasn’t been framed as a design problem. It’s been framed as a tools problem. Buy the right AI tool, plug it into the existing structure, carry on.

That framing is the trap.

Dorsey and Botha close with a question I think every leader should sit with: what does your organization understand that is genuinely hard to understand, and is that understanding getting deeper every day?

For marketing, I’d put it differently. What does your marketing organization know about your customer that no competitor could easily replicate? And is your structure designed to make that knowledge compound — or is it trapped inside the heads of a few people, locked behind reporting lines that were drawn in a different century?

The best performing marketing organizations I’ve seen weren’t the ones with the biggest budgets or even in some cases the best talent. They were the ones that were designed. Deliberately. For the world they actually operate in.

Most marketing orgs weren’t designed at all. They grew. Layer by layer. Reorg by reorg. Until the structure became the strategy — and nobody noticed.

Human led. AI powered.

frankiejeangroup.com

By slide five, I asked the same question each time: what have you actually decided differently because of this?

Silence.

Not one leader in those rooms could tell me what they’d stopped doing. What they’d accelerated. What judgment call the AI had informed. The technology was everywhere. The leadership was nowhere near it.

I don’t think those rooms were unusual. I think they were representative.

McKinsey published a report earlier this year that should have been uncomfortable reading for every C-suite in the world. They surveyed over 3,800 employees and executives and reached a blunt conclusion: employees are ready for AI — the biggest barrier to success is leadership.

Yep.

The bottleneck is not adoption. It is not the tools. It is not budget or technical skill. It is the people at the top.

C-suite leaders are more than twice as likely to blame employee readiness as they are to examine their own role. Meanwhile, employees are three times more likely than executives realize to believe AI could handle 30% of their tasks within a year. The workforce is already there. The corner office is still looking at slides.

BCG’s numbers are even sharper. Out of more than 1,250 global firms studied, just 5% are generating real returns from AI. Sixty percent report little to no benefit — minimal increases in revenue or cost savings despite substantial investment.

Five percent! After billions in spend. After years of pilots. After every conference keynote promising transformation.

So what separates the 5% from the rest?

It is not the model they chose. It is not the vendor. It is not the size of the budget.

The determining factor in AI success is no longer individual capability or even the technology itself — it is organizational structure, policy, and the way leaders choose to approach integration.

AI amplifies whatever is already there. Hand it to a well run team with clear strategic direction and it accelerates everything — research, synthesis, production, analysis. Hand it to a team without direction and it produces more confusion, faster, at greater scale, with better formatting.

McKinsey found that high performing AI organizations are three times more likely to have senior leaders who actively own and champion their AI strategy, not delegate it to a transformation office, not outsource it to IT, not wait for a vendor roadmap. They stay at the centre of the decisions.

That is the pattern. The organizations winning with AI are not the ones with the most aggressive adoption mandates. They are the ones where a senior leader defined what the machine does, what the human does, and held that line.

For CMOs, this should land harder than it does in most boardrooms.

Because the things that define great marketing leadership — vision, commercial judgment, creative direction, client relationships, accountability — none of these can be delegated to a model. An AI will not tell you when a strategy is technically correct but commercially wrong. It will not read the shift in a client relationship before the brief lands. It does not carry the consequence of the call.

The CMOs I see outperforming right now have done something deceptively simple. They mapped their team’s work and drew a line. Everything below the line — research, first drafts, data synthesis, reporting etc — goes to AI. Everything above it — the judgment, the direction, the relationships, the decisions that carry risk — stays with a human.

That split is not passive. It is a designed operating model. And it has to be deliberate, because in a world where any team can produce AI grade output, the quality of the human thinking on top is the only differentiator left.

But drawing the line is not enough if you don’t understand what’s on the other side of it.

The CMOs getting this right are not simply delegating AI to their teams and waiting for the output. They are in the tools themselves. They understand how the build process works — not at a top tier engineering level, but well enough to know what’s possible, what’s fast, and what’s a waste of time. They know the difference between a prompt that produces filler and one that produces a genuine first draft. The vibe code. They know when an AI workflow is saving their team twenty hours a week and when it’s just generating volume nobody asked for.

“Human led” does not mean human-as-before. The playbooks that built the last era of marketing — the instincts, the processes, briefing, reviewing, approving and executing — were designed for a world where every piece of output cost time and people. That world is gone. A CMO still running a 2019 operating rhythm with a 2026 technology stack is not leading. They are presiding.

The leaders who will define the next decade are the ones who can hold a board level strategic conversation in the morning and actually go build the solutions in the afternoon. Think and build. That is the new minimum, because you cannot draw a credible line between what AI should do and what a human must do unless you have working fluency in both.

BCG calls the 5% who are generating value “future built.”

Those companies say most of their AI value comes not from the tools themselves but from reshaping how work gets done. Not automating old processes. Redesigning them around a new division of labour between human and machine.

That is what “human led, AI powered” actually means. It is not a tagline. It is a specific operating model — the difference between AI producing better looking confusion and AI genuinely changing what a team is capable of.

Every organization using AI right now is producing more output. Very few are producing better decisions. The difference is always the human at the core — and whether that human evolved fast enough to earn their role.

AI made the deck. Someone still has to make the call.

frankiejeangroup.com