In early 2024, Big Tech’s escalating GPU arms race was still framed as an AI land grab where the means (capex and compute) mattered more than the ends (revenue and profit).

Fast-forward to the back half of 2026 and time is catching up. Meta is indeed having trouble figuring out how to turn a profit on consumer AI. This is why I posed my original question all the way back in May 2024: it’s a very difficult problem to solve!

While consumer AI may be a tough nut to crack, B2B AI may be more straightforward. You just sell shovels to gold miners via a cloud platform.

To this end, Meta still lacks a traditional hyperscale cloud offering like AWS, Google Cloud, or Microsoft Azure. Without a mechanism to sell AI capacity to other developers, Meta must rely solely on its in-house product and engineering teams to crack the problem of how to productize AI and make money from it.

It might seem that Meta is facing a high-stakes either / or choice:

1. Stay the course and mine for gold—Continue to gamble that Meta’s in-house teams can “solve” consumer AI before Meta goes broke; OR

2. Change course and sell shovels—Make a drastic pivot and compete with huge incumbents selling GPU capacity at a markup that will steadily erode.

This post makes the argument that —as a new hyperscale market entrant— Meta doesn’t need to adhere to 20-year-old cloud paradigms and has an opportunity to forge a third path that I’ll call the “B2B2C” AI cloud.

OpenAI and the Rise of the “Neocloud”

OpenAI serves as an example illustrating how “cloud” is an evolving, changing concept in the age of AI.

They probably understand that going toe to toe with legacy hyperscalers on raw cloud computing resources is a losing strategy. Instead of cloning their own unbundled menu of 200+ cloud primitives, OpenAI seems to be imagining something differentiated and more suited to the present-day state of the art.

OpenAI’s approach gives us a convenient preview of what certain aspects of an “AI-native cloud”—or neocloud—might look like.*

Among other things, a neocloud offers features to account for the fact that AI workloads are not procedural, deterministic, or linear. Stateful APIs, persistent memory runtimes, and agent orchestration are all examples of this.

For the record, OpenAI is not the only neocloud. Many other startups and companies have recognized the need for AI-native clouds.

* The legacy hyperscalers are not standing still and have their own versions of all this stuff.

Like OpenAI, Meta Needs to Avoid the Raw Compute Pricing Trap

If Meta enters the cloud arena, it must also avoid becoming a landlord renting out commodity raw compute at a markup.

Compute power is inherently subject to price erosion and hardware obsolescence. AWS, Azure, and GCP finance teams have had this dialed in for a long time; long before AI came along. And I am guessing that legacy hyperscalers are far better prepared than Meta to fight a long war of attrition on cloud pricing, especially for large customers.

Instead, a defensible AI cloud could be built on Meta’s real moats:

• Consumer apps, media, graphs, and data at the scale of billions of MAUs

• Global distribution footprint across apps like WhatsApp, Facebook, Instagram, and Threads

• Highly motivated, nimble, and tech-savvy global customer base of B2C brands and businesses, many with deep experience building and growing popular B2C apps and UX patterns

The Data & Distribution Footprint: Meta vs. OpenAI vs. AWS/GCP

When you further examine the cloud landscape through a B2C lens, the limitations of existing platforms become clear:

• AWS, Azure, and Google Cloud are “legacy” clouds in the AI era. They take a traditional approach that maintains strict technological and commercial firewalls between their cloud infrastructure, consumer datasets, and their owned & operated media (e.g., Google Search, Gmail, Amazon.com, Amazon Prime TV, Xbox, etc). Enterprise cloud customers can only access consumer signals through specialized, siloed clean rooms—like Amazon Marketing Cloud (AMC) or Google’s Ads Data Hub (ADH).**

• OpenAI is more of a true neocloud, offering features like stateful APIs. However, its consumer app and data ecosystem is still in its infancy compared to Meta. OpenAI cannot offer developers native access to rich, real-world consumer behavior or identity graphs.

** Cloud customers can access vast troves of 3P consumer data via cloud marketplaces, but this doesn’t constitute a truly native solution.

Traditionally, a business attempting to put a B2C AI application into the wild needs:

1. Raw compute from one place (a hyperscaler)

2. Distribution from somewhere else (typically an app store or ad platform).

The accepted hyperscaler model is definitely “find your own users.”

But does it necessarily have to be this way? What if Meta used its entry into the hyperscaler market to redefine the category, offering native consumer data and B2C distribution alongside an AI-native neocloud?

Enter the “B2B2C” cloud

Traditional clouds are B2B. The “2C” part is the customer’s problem. But Meta has “C” in the BILLIONS, man! Maybe that gap could be bridged!

A Meta B2B2C cloud would provide:

• The compute to run the app;

• The native consumer context to personalize it;

• A built-in distribution pipeline across Instagram, Facebook, WhatsApp, Threads and other apps to seamlessly reach billions of end users.

It’s worth pausing to explain why I’m emphasizing end consumers and the apps they use. It’s very clear that Mark Zuckerberg wants to put AI into the hands of every person on Earth. Read his most recent essay “The Future is for Everyone.” The faster Meta can accelerate distribution and consumer adoption, the better.

An Army of Millions: The B2C Marketer Ecosystem

There is another massive advantage Meta possesses that traditional hyperscalers lack: a deeply embedded global community of millions of B2C marketers and growth hackers.

AWS, GCP, and Azure built their empires by selling to IT departments and enterprise CTOs. Meta, by contrast, has a customer base of millions of savvy SMBs around the world who understand consumer demand, performance marketing, and digital funnel optimization.

This community isn’t composed of casual hobbyists; it is a global army of commercial operators who understand how to convert consumer attention into revenue.

By offering a B2B2C Cloud, Meta would hand a high-powered AI development and distribution toolkit directly to the very businesses best positioned to build, commercialize, and scale consumer AI experiences.

The B2B2C Advantage: App Store Economics

By combining cloud compute with native distribution, Meta could change the economics of cloud computing:

1. App Store Revenue Share: Traditional cloud platforms charge by the gigabyte or GPU hour. App stores, however, take a percentage cut of top-line developer revenue. By offering both the computing backbone and the distribution funnel, Meta could capture higher-margin revenue-sharing instead of competing in a price war over raw infrastructure.

2. Crowdsourced Discovery Engine: Rather than relying exclusively on Mark Zuckerberg’s internal product teams to invent the next consumer “killer app,” Meta Cloud could enlist millions of independent builders to experiment using Meta’s AI cloud infrastructure. When a third-party developer hits gold, the breakout app runs on Meta’s compute, utilizes Meta’s signals, and generates platform revenue for Meta’s ecosystem.

Conclusion

There isn’t a neat bow I can tie on all of this. It’s all totally hypothetical at this point. But we do have kind of a unique situation. The next year or two are VERY high-stakes for Meta. They might do some drastic stuff. Note that I didn’t say “desperate.” I think that leasing raw compute at a markup might come across as desperate (I think Zuck knows this). Whereas reinventing the hyperscaler category to advance consumer AI could be an interesting move that would have Meta writing its own cloud narrative versus trying to play catch-up with the established giants.

Well, look who’s reinventing cable TV.

It will be fun to watch the ad industry pundit class all simultaneously take off their “Linear TV is a Dinosaur that Deserves to Die” hats and put on their “Linear TV is Actually Super Smart” hats. They can use this article as a quick reference for their talking points.

Indeed, there was a time around 2019 to 2021 when there was serious talk that no one would ever watch ads on TV ever again…I guess because Digital.

I’m painting with a very broad brush here, but Digital people tend to be technologists and technologists simply don’t understand advertising and scoff at the entire idea every chance they’re given. It never fails. Here’s my related rant on Elon Musk, advertising, and Tesla. But I digress.

It was easy to predict in the 2019-2021 timeframe that this TV advertising doomerism was all a fad. Streaming was new and Disney+ had just launched to much fanfare, so —as usual [eye roll]— everyone got distracted by the newness and, in their excitement, forgot a bunch of basic economic and pricing principles. Forget the high-minded advertising and media theories; we’re just talking money and arithmetic here.

Here’s my reaction to that same Digiday article the very next day:

Why was I so confident about this almost 7 years ago? Read on and I’ll gladly explain! Frankly I’m surprised it’s taken so long for the reinvention of basic cable TV to take place. Trust me: more platforms are going to follow in the steps of Netflix because their ad sales teams will be leaving money on the table if they don’t.

Yes, this article is me doing a sad little victory lap over some sh*t I tweeted in 2019, but then again I think there are a bunch of factors that just aren’t very well understood and are worth explaining.

Here are the reasons that it was inevitable that Netflix would reinvent basic cable. I’ll dig into each in more detail:

1. Netflix needs to maximize its Price-Quality Frontier; a curve representing what all potential Netflix customers are willing to pay for in return for ad-free viewing and other amities.

2. On-demand binge-watching is economic poison for streamers. Netflix and other streamers will naturally push their customers away from it over time.

3. TV “channels” are a great way to repackage content for your ad sales team. This is an old trick in the content monetization and publishing game: curate thematic sub-collections of your existing content and sell it as a new SKU. Magic!

4. Choice fatigue is real when it comes to streaming libraries. People like stuff they can watch passively. TV is a means of relaxation. Channels play into this.

5. Generative AI and content saturation will continue putting a premium on live viewing and simultaneously erode the value of on-demand libraries.

Monetizing expensive content libraries by maximizing the Price-Quality Frontier

Not everyone is willing to pay the same price for a thing. This is an old idea. It’s why —for example— General Motors declared in 1925 that the company would offer “A car for every purse and purpose.”

Lots of people want cars, but they’re not all willing to pay a single, fixed price. Some people will pay more for a car, others less.

General Motors’ answer to this was to maintain five separate car brands, roughly grouped into price tiers, with Chevy being the most affordable and Cadillac being the most luxurious.

This is also true for entertainment. Not everyone is willing to pay the same amount for a streaming subscription. Some people don’t want to pay anything at all.

In exchange for flexibility on price, people will make trade-offs. For example:

• Watching ads

• Delayed access to new releases

• Limited content libraries

• Limited account sharing

So it holds true that streamers that only charge one price are sacrificing the customers that aren’t willing to pay that much. A lower-priced tier with trade-offs will capture some of those potential subscribers. An even lower-priced tier with even more trade-offs will capture more incremental subscribers. And so forth.

We can plot this as a “Price-Quality Frontier”—a curve that represents what people are willing to pay for at what price points, and what trade-offs they’re willing to make in return for discounts.

And guess what? No matter what you charge or how many subscribers you have, your content library still cost you a fortune to produce or license in the first place. Movies and professionally-produced shows are expensive and represent a very high fixed cost. Therefore it’s absolutely critical for content owners to find the most efficient possible combination of monetization options that collectively maximize revenue and profit.

This means that a streamer like Netflix will only maximize its revenue when it can monetize as many zones of the Price-Quality Frontier as possible. The most efficient and revenue-maximizing approach will push the boundaries on the trade-offs its viewers are willing to make (or the amenities they expect) in exchange for flexibility on price.

The obvious way to monetize the potential subscribers who don’t want to pay anything for Netflix content is to have sponsors and advertisers subsidize the missing subscription fees and show ads to viewers.

However, Netflix already has a paid tier with ads. So what to do?

A totally free tier will probably have more ads crammed into the viewing experience, but another variable Netflix can play with is user choice. This is a brilliant aspect of linear TV and the concept of “channels”: it removes some (but not all!) user choice in a format that is generally liked by people who want free TV programming. Multiple theme-based “channels” give viewers some control and choice.

And to think this was all invented decades ago in the previous century!

Binge-watching is economic poison for streamers and content owners

Allowing subscribers to binge through all-you-can-eat catalogs of expensively-produced shows and movies won’t last forever. It makes very little long-term business sense.

There are really two factors at play:

1 - Maximizing the lifespan of entertainment IP

Binging cuts deeply into the lifespan and emotional/cultural impact of entertainment IP. Entertainment content has very, very high up-front costs. It’s just part of the business. Before you’ve sold a single subscription, ticket, song, book, licensing deal, or anything else, you’re already deep in the red. You don’t make money by constantly producing new IP because most movies and shows will fail to turn a profit; you make money by milking IP that’s already proven popular and throws off higher and higher profit margins over the years. And hit properties like Star Wars and Seinfeld, James Bond and Jeopardy! come along pretty rarely.

Binge-watching could not be more anathema to the entertainment industry’s need for disciplined multi-year planning, long IP lifespans, and long-term emotional investment from audiences. When a fickle streaming subscriber binge-watches a show in a weekend and then moves on, the streamer captured a negligible amount of low-margin (or even negative-margin) subscription revenue while at the same time sacrificing years of future high-margin engagement.

I could write a whole post on this, but I’ll stop here. Binge-watching is a consumption pattern that is at odds with the production economics. It’s utterly unsustainable for streamers and the entertainment industry.

2 - Minimizing instant gratification and reducing subscriber churn

Related to the above, binge-watching simply means that subscribers will exhaust the value of the streaming platform catalog faster. And the faster the value dries up for a given subscriber, the faster they will churn.

Linear and appointment-based TV is pretty brilliant in this respect: it always gives viewers a reason to keep coming back. Just because it’s technically possible to now offer complete video catalogs for on-demand, all-you-can-eat consumption doesn’t mean it makes business sense.

Even keeping subscribers an extra few weeks on average would make a massive financial difference to any streamer. This means that streamers must start holding things back. Many have already pivoted to models where new episodes of shows are dripped out weekly. A logical next step is to launch linear “channels” that nurture appointment-based viewing relationships that last longer (and develop a deeper emotional connection) than binge-watching.

TV “channels” are a great way to repackage content for your ad sales team

A TV channel is a curated set of entertainment content. Curation adds value in a number of ways. This value translates to easier ad sales and/or higher premiums paid by advertisers.

• Theme- or demographic based TV channels mean engaged, targeted, segmented viewers. Advertisers love this. If I have to explain why, you’re reading the wrong blog.

• From a purely subjective standpoint, curation can add a lot of “sizzle” to an otherwise static content library. You can create exciting branded channels like Netflix ACTION, Netflix HEIST, Netflix BABY, Netflix COOKING, or Netflix BRITISH. You can re-use the same content across multiple channels and magically create value because you packaged it and branded it.

• Free TV channels give advertisers a way to target net-new eyeballs and add incremental reach. Potentially this would include viewers that advertisers cannot target on paid ad-free streaming platforms. For example, a household might be ad-free on Disney+ but might be perfectly happy to watch ad-supported free Netflix channels. This is a win for Netflix ad sellers.

• From a very tactical perspective, if you’re selling ads and sponsorships, it never hurts to have new SKUs to talk to advertisers about. Repackaging existing content into new SKUs is a very old page out of the content monetization playbook.

I’ll pause while a few of you copy/paste those bullets into an email to your Board.

And while I’m on the subject: I truly expect YouTube to adopt this approach and launch curated linear channels using content from its top creators. Given that the majority of YouTube viewing now happens on televisions, it makes a ton of sense and would allow YouTube to create new premium TV ad inventory practically out of thin air.

But look: TV invented all this decades ago in the previous century. I’m not saying anything new. There were probably radio people in the 1950s ranting about how the TV people were stealing their ideas from the 1920s.

Choice fatigue is real

Just because you can give viewers access to an exhaustive on-demand content library doesn’t mean you should.

On-demand course catalogs are trying to force an active UX into an experience that is ideal when it’s passive. When people watch TV, they want to settle back into their couch and relax. This mis-match is visible in all kinds of statistics.

There are a number of downsides for streaming platforms:

• Decreased emotional engagement from viewers who are overly pre-occupied with navigating the UI. This has an intangible but real long-term financial impact. Emotional connection is the profit engine of the entertainment business.

• The erosion of perceived platform value. This will be largely subliminal, but it does play a measurable role in how consumers spend their streaming time and money. A 2024 study found that about half of people strongly consider a streaming platform’s UI when deciding whether to subscribe. Many platforms have addressed this by making smaller and smaller marginal improvements to their recommendation algorithms. TV channels are a much lower-tech and simpler solution to the same problem. 🤷‍♂️

• Viewers who are endlessly browsing aren’t watching ads. OK maybe they’re seeing a homescreen ad here or there, but in general if you’re trying to monetize content, you want people watching the content. Every extra minute viewers are spending in “browse mode” comes straight out of a streamer’s bottom line.

TV channels are a great compromise. Viewers have some control: they can change the channel. But the channel itself just has something “on.” You start watching and within seconds you’re nicely zoned out. If you think about what TV viewers want from entertainment, this is an ideal, relaxing experience.

Choice is a mirage anyway. Not only do people not mind having choice taken away from them, they actually prefer it! Ask any good UX designer. This is another topic you could write a whole book about, but I’ll move on.

Generative AI and content saturation will continue putting a premium on live TV viewing

This last point is more thematic and unpacks in too many directions to fully cover here. For example, I wrote about live, shared experiences with respect to AI in this April 2026 post on OpenAI’s acquisition of TBPN.

Generative AI means that the cost of content production will generally get lower. This means that the barrier to offering a content catalog (a bank of shows and movies) will get lower. Content catalogs will get gradually more commodified. This means that there will be downward pressure on the amount people will pay to access content catalogs, and the rates (CPMs, etc) advertisers will pay as the supply of available content continues to balloon.

This is already happening as streaming platforms have unleashed thousands upon thousands of hours of on-demand content over the past decade to a fickle viewing public with little loyalty to any one platform.

There are many other variables at play and many counter-moves content owners can make. But generative AI will add to this kind of downward pressure over time.

A key variable for content owners to control is 'live-ness' or the degree of synchronicity required to enjoy a piece of media. By dictating which content must be consumed at an appointed time versus what can wait, platforms can dictate the urgency and attention a viewer must bring to the screen.

This has already been happening with live sports. This is likely unrelated to AI, but it’s a “teachable moment” on the value of engaging live content.

Here’s the thing: sports is only one type of content and not everyone watches sports.

But the shared cultural moments of live sports viewing are near-priceless in the “post-monoculture” 21st century. I write this in the midst of the 2026 World Cup which has, wow, been a global cultural and entertainment touchstone in a way we haven’t experienced for many years now. I’m not even a soccer fan and I’m vicariously basking in the glow. Ditto the Knicks vs. Spurs in the NBA finals. I don’t follow basketball and even I was inundated with awesome fun basketball content.

So how do content owners and streaming platforms cultivate shared experiences and cultural moments for all the people who aren’t sports fans?

In terms of programming, that’s a topic for another blog post. But the form factor needs to be linear. Linear TV channels are a great option here as they offer an always-on formula for live, appointment-based viewing. The alternative looks more like pay-per-view where streamers would have to create urgency around live events one by one. This can pay off for a small handful of intensely-hyped events (e.g., one or two per year), but it’s not an efficient, repeatable approach if the goal is to build a steady, predictable stream of ad revenue.

That’s all for now

Well that turned out to be a very long post! I hope you enjoyed it. A lot more could be written on this topic. I didn’t really have the time or energy to get a little more "mathematic” and model additional efficiency curves, content payback timelines, and so forth. But my point from the introduction stands: some of this is about the entertainment & advertising business, but a lot of it is based on generalized principles of pricing efficiency and payback on high fixed costs.

Bye!

LLMs and AI agents might mean that the “grand bargain” of web search is dying, but it was a very powerful virtuous cycle while it lasted. It worked like this:

1. Google makes a search engine that works far, far better than anything else available at the time.

2. You, a website owner, allow Google to scrape and index your site.

3. With this data, Google will feature your site in “organic” search results. Your site data also makes the whole search engine work better, which will drive yet more consumer adoption.

4. Google will also sell “paid” search results alongside the organic ones.

5. Either way, Google has opened a conduit for driving traffic to your site, which (let’s face it) would be otherwise unfindable in the unending mediocrity of the internet.

PS: you, a website owner, can also monetize traffic from Google by selling ads using some other tech that Google will also happily provide. 🤷‍♂️

A virtuous cycle can only gain momentum to the extent that its participants can actually perceive its benefits.

The virtuous cycle of that “grand bargain” was made tangible to website owners by Google Analytics, which launched in 2005 as a free service.

This was a really big deal. Access to Google Analytics caused pretty much the entire world of publishing, entertainment, travel, B2B, etc completely ADDICTED to search traffic and eager to keep feeding into the virtuous cycle.

The virtuous cycle of web search turned Google into the greatest money-printing advertising machine ever known.

I am going somewhere with all this exposition, I promise.

At the risk of stating the obvious, as the virtuous cycle got stronger and Google search traffic grew and grew, Google was able to print simply ridiculous amounts of money from search ads.

It could be argued that without Google Analytics, history would have turned out differently. Google search ads would likely have been quite profitable, but perhaps more modestly so. Google Analytics gave the rest of the world skin in Google’s game.

In short: Google Analytics → compounding virtuous cycle → insane ad revenue

ChatGPT doesn’t have a virtuous cycle and this is a problem.

This applies to any LLM, but I’m going to pick on ChatGPT because they badly need to stand up a cash-cow ads business.

This means that ChatGPT probably needs to establish a virtuous cycle or “grand bargain” of its own.

However, there’s a big obstacle to establishing this new “grand bargain:” no one can tell what the f**k happens inside an LLM. It’s been difficult for companies to get skin in the LLM game because you feed your data in and get practically zero insight into what happens after that.

Even LLM designers don’t fully know how LLMs work. I don’t say this with any sarcasm, just to reinforce that LLM opacity is a legitimate phenomenon.

Enter the GEO platforms…

To fill the gap, a slew of tech vendors have cropped up. Companies like Profound, Evertune, and Bluefish can help brands, publishers, and other companies understand how (and whether) they show up in LLM chats.

“Generative Engine Optimization” (GEO) is the goal and the name of the software category: to understand and then optimize how a brand, product, or service shows up in LLM chats.

My understanding is that GEO platforms take basically a brute-force approach:

If I am the Toyota RAV4 marketing team and I want to know what happens when car shoppers ask ChatGPT about crossover SUVs, then my GEO platform will hit ChatGPT’s API with a slew of various prompts to build up a statistical picture of what comes back. A “slew” could likely mean thousands upon thousands of prompts like:

• “what is the best crossover suv?”

• “what is the best crossover suv for families with three kids?”

• “what is the best crossover suv for families with three kids and a dog?”

(and those bajillion prompt premutations need to be repeated daily or weekly or monthly or whatever)

You can see how this gets ridiculous and why I characterize the approach as “brute force.” But it’s probably the best option right now. Because brands can’t directly snoop on people’s chats and…

[…drumroll…] 🥁

…ChatGPT doesn’t provide its own native analytics platform for marketers to analyze what’s going on inside ChatGPT.

Google Analytics for ChatGPT

Well, if we’re splitting hairs or whatever, we’re really talking about Google Analytics and Google Search Console. I’m just going to say “Google Analytics” to cover both, in part because no one knows how this might work for LLMs. Would this be two separate tools? One? Three? Who knows.

I don’t see how ChatGPT proceeds with its ads strategy without ChatGPT Analytics or whatever they’ll end up calling it. This post is a prediction that OpenAI will launch ChatGPT Analytics fairly soon. I wouldn’t be surprised if we see it by the end of 2026. ChatGPT’s ads strategy has been bumpy but progressing quickly now that it’s live.

But what will ChatGPT Analytics look like? What will it measure?

Basic stuff to start with, I suppose. Keywords and phrases, maybe some sentiment analysis. Chat length. Basic intent categorization (travel, health, advice, etc). I imagine one could look at existing GEO tools and get a good idea of what the ChatGPT Analytics MVP will offer.

However, given the mercurial nature of LLMs, measurability and observability will probably be difficult. Web search is a structured list of ranked links. The link text comes from the publisher’s meta tags. It’s all very orderly and standardized. LLMs are currently the exact opposite; a blank and chaotic canvas.

How might ChatGPT Analytics work?

This is kind of the fun part. LLMs and web search don’t work the same way, so we get to put on our thinking caps and imagine what kind of features ChatGPT Analytics might have.

Here are a few ideas:

• Standardized UX Patterns and Chat Telemetry: Currently, every LLM output is too unique of a snowflake to run standardized “apples to apples” analytics on. To fix this, ChatGPT would have to lean into structured UI components—such as interactive widgets, specialized app plugins, or clear citation blocks. A native analytics platform would track chat telemetry, possibly analyzing the exact point at which a user ends a chat. Did they exit because the AI fully satisfied their intent, or did they bounce out of frustration? Boredom? Understanding the termination point of a user journey provides insight into whether they got what they were looking for.

• Deep Intent Categorization: Publishers, content owners, brands, and creators need insight into the high-level user intents driving the system. If ChatGPT can categorize and feed back data showing that a creator’s content consistently surfaces during “high-intent” commercial tasks versus “low-intent” casual queries, it provides a crucial signal. If a publisher discovers their content is surfacing in areas they cannot monetize, it tells them how to pivot to greener pastures.

• Native training Data Provenance: While full real-time training data attribution is computationally intensive, a native tool could offer periodic, scaled metrics and telemetry regarding content lineage. These analytics could come from inside the LLM. Analytics could show a content owner how deeply their content or data persisted through various training phases—answering questions like: Did this piece of content fundamentally add to the unique knowledge base of the model, or was it filtered out as repetitive noise?

Here are those same three pillars put differently:

• Standardized UX patterns → gives ChatGPT, content owners, and advertisers standardized, discrete elements to measure

• Intent categorization (commercial / casual) → gives content owners and advertisers a way to understand where to spend their time and a way for ChatGPT to make its LLM less extractive and more additive for content owners

• Native training data provenance → gives content owners a feedback loop to understand if they’re doing “SEO for LLMs” right and incentivizes everyone to feed the highest possible quality content and data into ChatGPT

How do you think ChatGPT Analytics will work?

Leave your ideas in the comments. How will ChatGPT Analytics work? I know there are lots of people walking around out there who know way more about this opportunity than I do (which is admittedly little). From a purely strategic perspective I’m convinced that ChatGPT will do this, but the precise features of an LLM analytics product are more of a guessing game.

Thanks for reading. 🙏

I agree, more or less.

This post isn’t a justification for the acquisition. I have no idea what OpenAI’s rationale was. This post is where I landed when I pondered whether there might be deeper reasons or lessons behind it. Like, if you really navel-gaze over this acquisition, what conclusions might you draw?

AI is Boring

As the honeymoon phase of the LLM era ends, we’re left with a nagging question: why should anyone—outside of AI-pilled prosumers and the tech community—actually care about AI on a mass-market scale?

And why would one of the leading AI platforms acquire a media property that models itself after television, the dominant media from the previous century:

• TBPN doesn’t use AI to generate video. Just video cameras pointed at real people. AI aside, we’re talking pretty low production values.

• TBPN is broadcast live. TBPN is very much meant to be watched in real time. You are supposed to watch it between 11am and 2pm PST, the only time window in which it airs. How quaint!

To the uninitiated, buying a quasi-television media property looks like a distraction for a company building AGI. But if you look at the history of technology and media, OpenAI is making an admission: AI currently has no allure.

Boring Things Don’t Command Big Ad Dollars

Think about that, then think of all the ad dollars OpenAI will need to take from Google, Meta, and an army of other competitors.

There is a massive opportunity cost to being boring, at least if you’re chasing the B2C market.

(I realize that OpenAI may be prioritizing B2B business these days, but f**k they change their strategy every 5 minutes, so they really do need to keep their options open.)

The Lonely LLM

Right now, AI is a lonely experience. When you prompt a model, you are in a private, atomized bubble. If an LLM tells you something brilliant in response to your equally brilliant prompting, there is —tragically— no one else there to react to it. AI’s value is only realized when you brag about it on LinkedIn or X to other AI dorks.

This is the opposite of how mass media creates value. Think about the Super Bowl or the series finale of Cheers. TV programs aren’t just content; they can be shared experiences on a national scale. They have value because millions of people watch, react, and validate the experience simultaneously.

We could look at newer media too. Why is TikTok so fun? It feels like a shared experience. The likes, the comments, the follows, the DMs. All of these are ingredients that help each TikTok user share their excitement or disgust or whatever with millions of other people worldwide.

Tech nerds call this a “network effect.” To each individual user, the value of a media platform grows proportionally as more users also consume that media. Ad dollars follow. Nothing succeeds like success.

In order for AI to capture the kind of emotional investment that other forms of ad-supported media enjoy, it needs to find its way out of the chat window, out of a hobbyist context, and onto a stage. No normal person will give a rat’s ass about AI until it becomes a spectator sport.

Escaping the Confines of the Hobbyist Crowd

Present-day AI is reminiscent of 1970s hobbyists building their own desktop computers from Radio Shack kits. So many passionate, autodidactic builders. Pioneers and trailblazers! That kind of thing is a dud for pop culture, however.

Microsoft famously broke the shackles of desktop computing’s early hobbyist / technologist years with its launch of Windows 95 (in, um, 1995). That launch required a massive advertising and PR campaign. And it allowed Microsoft to make a huge leap in bringing computers into the popular imagination.

Now…did OpenAI really need to acquire TBPN outright as its own in-house publicity arm? I would argue not, but the acquisition implies that OpenAI —in its quest for ad dollars— at least realizes which game it’s playing (and it’s not a technological one).

It’s the Glamour, Stupid

There is also the matter of glamour and fame—two things that an LLM cannot generate out of ones and zeros. We are biologically wired to bask in the glow of human celebrity. This is a key limitation of AI. By definition there isn’t a human on the other side! Therefore glamour and fame are absolutely unattainable. Pretty bad for AI.

OpenAI may recognize that while their tech is powerful, it is sterile. With its parade of high-profile guests, TBPN provides a celebrity halo effect. OpenAI can wrap itself in the trappings of professional sports-style and cable-news-style TV coverage and live interviews with Important People Saying Important Things.

This is an area where it makes little sense that TBPN got rid of its sponsors, though. Economically, TBPN may not need sponsors as part of OpenAI, but the sponsors are part of the spectacle! When a thing is sponsored, the stakes are raised. There is something much more compelling about race cars or soccer jerseys covered in sponsor logos than plain ones.

Pop Culture Sea Legs

Is this all a sign that OpenAI is getting realistic about AI’s limitations? Perhaps. It shows they understand that faster AI models won’t spark the kind of desire and mass appeal that commands real ad dollars. If OpenAI believed that AI could emulate these qualities, they’d have acquired some vapid generative AI video platform. But they didn’t. They acquired two guys and some video cameras.

OpenAI may be using TBPN to get their “sea legs” in popular culture and the realities of ad-supported media. TBPN knows how to talk to humans and create shared experience while still letting OpenAI stay in its safe space of terminally-online tech culture. For now.

I won’t be surprised if they use the muscles they’re developing with TBPN to make forays into live sports and find ways to integrate AI into the fan experience. Just a random prediction.

I’m done navel-gazing for now. Thanks for reading.

A few extra thoughts on AI headwinds versus tailwinds

From the “Agentic Commerce Won’t Amount to Much” camp, so much of the analysis is focused on adding up the many headwinds that AI apps and UX patterns will face as they attempt to displace pre-AI apps and UX patterns.

Almost the entire economy (with perhaps the notable exceptions of companies like OpenAI and NVIDIA) is a pre-AI incumbent. Every major app, site, and device is still built on (and makes its money from) pre-AI UX and consumption patterns. These incumbents will not be easily displaced and they will want AI innovation to happen at a pace they control.

So it’s a little hard to judge the success of agentic commerce when AI is still in its clunky infancy and all we’ve seen to-date are headwinds. Headwinds are the default state.

Tailwinds, especially those driven by grassroots consumer behavior, are harder to predict, but they often aren’t subtle.

CD sales peaked around the year 2000, almost exactly when Napster broke out with a superior UX and consumption pattern (from the music listener’s POV, not the incumbent recording industry's). This was not subtle. People love instant gratification and Napster was revolutionary. I remember the first time I encountered Napster (probably around the year 2000). I was truly blown away and delighted by the idea that you could just look up any song ever recorded and start listening to it within a few seconds.

Ditto 2007 when Steve Jobs made every BlackBerry with a keyboard look like geriatric garbage overnight. The glass smartphone touchscreen was a new and magical UX pattern that people loved. Smartphones literally reshaped the world as those little glass screens in our pockets became “remote controls for our lives.”

Other than ChatGPT, AI hasn’t yet had a “breakout” consumer app or new UX pattern (other than that one man’s voiceover voice that’s now in every direct-response video ad). But I am confident it will.

When these new AI-driven UX patterns emerge, they will come as a breath of fresh air to shoppers and will make pre-AI UX patterns seems anachronistic.

From the inverse perspective, it’s difficult to imagine a future where all the time-consuming, cognitively taxing manual bullshit of shopping online (or using a keyboard or touchscreen for practically anything, for that matter) just persists indefinitely on autopilot. I really scratch my head at that idea.

Once a strong consumer adoption tailwind picks up, it may simply be unstoppable. Many once-legitimate agentic commerce headwinds will become moot as everyday people’s consumption patterns simply go somewhere else and organically find a path of least resistance.

Obviously what I’m saying is speculative. I can’t point to an agentic commerce “killer app” because we haven’t really seen it yet. But given the speed at which new apps and new UX/consumption patterns now emerge, it could happen quickly. TikTok went from not being on anyone’s radar to being a US national security concern in like five years? ChatGPT reached 100MM users in two months?

When everyday people find something they like, it can catch like wildfire. I think it’s an externality that a lot of agentic commerce analysis isn’t accounting for.

If you haven’t checked it out yet, here’s that Marketecture post again.

Here’s that post if you want to check it out, otherwise keep reading further down.

Turns out OpenAI may have the same idea:

Did Sam Altman just announce an OpenAI cloud service?

This is admittedly a bit different than what I wrote about in 2024. Meta sits in a specific industry sector as a social media and ads platform, whereas OpenAI wants to be an industry-agnostic software and technology company like Microsoft.

Therefore, an OpenAI cloud platform would probably look different than one from Meta.

But it’s still worth thinking about what an “AI-native” cloud platform could look like and why OpenAI would want to build one.

• A more focused and specialized AI cloud service: OpenAI wouldn’t necessarily need to clone AWS, GCP, Azure, and the like. A focus on AI-based applications would allow OpenAI to trim a lot of the bloat and complexity that the major clouds have accumulated over the years.

• More chances to strike AI gold: like Meta, OpenAI would likely use its own cloud platform to nurture potentially hundreds of thousands of devs and entrepreneurs around the world to uncover new “killer apps” for AI. To-date, while LLMs have been a game-changer, it’s still hard to point at one or two AI applications that are clearly beneficial and exciting to everyday people. AI is still in its infancy and is too esoteric to be both simple and revolutionary in the way that email was in the 90s or the iPhone was in the 2000s. This has led to a lot of frustration, disillusionment, and cries of “bubble!” in the tech industry. Companies like OpenAI and Meta badly need to solve this problem before they run out of cash or the market implodes.

• Make Zuckerberg’s AI capex gamble look foolish: let’s say OpenAI launches a cloud platform in Q1 of 2026. Sam Altman suddenly has a more positive story for investors and the marketplace. With a cloud platform that directly monetizes its expensive GPU capacity, OpenAI can now hedge its bets and sell shovels to gold miners just like the grownups at AWS, Azure, GCP, and Oracle. Mark Zuckerberg’s “I’d rather blow a trillion dollars than miss the boat” strategy now looks overly risky by comparison, as he has placed only one bet which is to dig the gold up himself. In my May 2024 post I go into this quite a bit more. But I could foresee Zuckerberg suddenly being under even more pressure than he already is to turn AI into a money-maker pronto.

It certainly is interesting to watch this play out. Personally I hope these guys crack the code before the market tanks.

I don’t consider it a question of “if” AI will revolutionize software, but “when.” And if you’re in the “when” camp as I assume Sam Altman and Mark Zuckerberg also are, you HAVE to consider ways to de-risk and set yourself up to ride things out. To that end, renting GPU capacity via a cloud platform is one of the more obvious options.

Thanks for reading. Hopefully I’ll post again sooner rather than later.

— Mark Zuckerberg

That’s the quote from the Q1 2025 Meta earnings call. You could write that quote off as naive or just bombastic, but at least Zuck is connecting AI to something that’s actually measurable: GDP. And it is true that big technological changes sometimes alter the mix of GDP.

It got me thinking: given that Zuckerberg is talking about something that is measurable on a macro scale, how exactly might one validate or invalidate what he’s suggesting?

If you read to the end of this post, you’ll find that there are indeed some threads to pull on. But first we’ll start with a history lesson.

The Economic History of Advertising

Historically (say, back to the 1920s), advertising has accounted for around 2% of US GDP. It’s notably consistent, given that AI isn’t the first wave of change to sweep through the advertising and media world during the last hundred years.

The 1920s saw the explosion of radio, the first real-time broadcast media in all of human history (seriously, think about that). The 1920s and 30s saw the emergence of “talkies”—movies with sound. The 1950s saw the explosion of television. The 1970s and 80s saw the explosion of desktop computing. The 1990s and 2000s saw the explosion of digital advertising and mobile devices.

It’s been a busy century!

But advertising dates back further than the 1920s. Advertising as we tend to know it emerged in the 1800s.

But why did it emerge? Why does advertising even exist? These questions are more than just navel-gazing if we are to predict how AI might fundamentally alter advertising’s role in the economy (as measured by its share of GDP).

Why Advertising Exists

In the 1800s, the Industrial Revolution brought sweeping change to the US economy.

With industrialization, two things happened:

1. Supply exploded: factories could now produce cheaper goods far in excess of a pre-industrialized “natural” state of demand. Instead of a local cobbler hand-making a few shoes for his fellow townspeople, there was now a shoe factory that could inexpensively produce hundreds of shoes a day.

2. Demand exploded: an industrialized economy generates surpluses beyond subsistence needs. More and more everyday people find themselves with surplus time and money. For the first time in history, “discretionary income” becomes a thing.

Factories are also expensive to build. Industrialized firms were under pressure to earn back investment capital and sell inventories of finished goods that piled up fast.

Enter advertising: a means for industrialized producers of goods and services to “artificially” generate demand, pitch their wares to more potential buyers at once, and accelerate sales (and minimize inventory turnover times).

The way I see it, advertising is an inevitable byproduct of these four factors:

1. An industrialized economy

2. Free-market capitalism

3. Free speech

4. Mass media

Advertising can be viewed as one of several mechanisms by which excess goods and services can be allocated among people and businesses that have excess money to spend and can exercise choice in how they spend it.

Advertising is an economic “matching” mechanism. But it’s not the only one.

How the Economy Helps Supply Meet Demand

The US economy is incredibly effective at matching supply (sellers) with demand (buyers). It has several mechanisms that aid in this process.

And when I say “supply” and “demand” in this context, I am generally thinking of finished goods and services, not raw materials.

I wanted to think of all the related mechanisms that could be measured or estimated as shares of GDP. Here the six tangible mechanisms I came up with:

1. Advertising—making demand aware of supply

2. Credit and finance—bridging the gap between the desire for goods and services and the immediate availability of funds

3. Distribution and retail—giving demand somewhere to go and buy supply; ensuring supply is feasibly and conveniently reachable by demand (which could also happen via delivery)

4. Trade—the ability for supply to move freely across regions, borders, and institutions in order to meet demand

5. Mass media, communication, and information dissemination—forms of communication or published material whereby demand can readily gather information on supply or consummate transactions (this would be measurable via industries like entertainment and publishing)

6. Sales and customer service—helping demand assimilate information, consider options, and make a purchase

(If you’re wondering why I left out Markets, it’s because markets and exchanges such as stock exchanges function as “matching” mechanisms for capital or raw materials, not finished goods. When you’re hungry you buy bacon from a supermarket, not pork belly futures from an exchange in Chicago.)

Why Is Advertising So Stable As A Percent of GDP?

If there are actually six different ways to help supply and demand come together, then why has advertising been so stable for last 100 years? Why does it hover around 2% of GDP?

If Mark Zuckerberg thinks Meta’s AI-based ad platforms will boost that number to 3% or 4% or beyond, it makes sense to understand why it’s been so stable up til now.

There are a lot of reasons, but these are the three that are most irreducible:

1. There is a human limit on how much advertising can be absorbed or even noticed (consciously or otherwise). We don’t know exactly what this limit is, but it certainly does exist, and it puts a constraint on advertising.

2. At a company level, there are diminishing returns beyond certain levels of ad spend where there is simply no more latent demand left to unlock, or defending against competition becomes more expensive (on a net, all-in basis) than doing nothing. (This would also equate to “buyer burnout” where buyers simply cannot and will not buy everything that’s advertised to them, rendering most ads useless past a certain point)

3. Mass media has a finite “carrying capacity” for ad messages. Firstly, there are only so many hours of media consumption in a day, placing some upper limit on just how much media can viably exist. Secondly, that media can only contain so many ad messages before it becomes unsatisfying to people who then turn away from overly ad-saturated media.

These three together represent the buyers, the sellers, and mass media communication throughput. All three components have inherent limits which together constrain advertising to around 2% of GDP.

Anything more than 2% (roughly, depending on the year) would essentially be a waste of capital, as it wouldn’t spark any incremental macro demand. Anything less would be wasteful elsewhere in the economy, as macro demand would fail to reach its full potential.

Could AI Shift the Balance?

Before we continue, let’s remember that share of GDP is a zero-sum game. Any increase in advertising’s share of GDP must come at the expense of something else.

This is a common fallacy that the “performance advertising” crowd falls into. There is an assumption that intrinsic improvements to ad effectiveness (e.g., hyper-targeted, zero-waste ads) will have some extrinsic effect too. This is far from true. Advertising has been getting steadily more efficient and innovative for the last 100 years and yet these gains haven’t caused any extrinsic effects where other parts of the economy have shrunk in favor of a bigger ad industry. This is why I pointed out three inherent macro limitations on advertising. More efficiency or “effectiveness” doesn’t change these limitations.

We start to see how Zuckerberg may be naive in his thinking, but let’s press on.

If AI is going to shift the balance and grow the ad industry on a relative basis, where will it take GDP from? That is the question we must answer to see if there is any substance to Zuckerberg’s statement.

This is why I laid out the six macroeconomic, GDP-measurable mechanisms that help “match” supply and demand for finished goods and services:

1. Advertising

2. Credit and finance

3. Distribution and retail

4. Trade

5. Mass media, communication, and information dissemination

6. Sales and customer service

We can, of course, disregard the first one.

So which of the other five might lose share of GDP in a world where Mark Zuckerberg is right? Which ones might shrink on a relative basis due to AI? (and keep in mind that my personal hope is that AI helps grow the economy on an absolute basis, so a component might shrink on a relative basis but still manage to grow on an absolute basis)

Credit and finance — NOPE

I’m not seeing a connection. Yes, AI will impact all five components and make things generally “better” but I can’t begin to think of why AI would cause Advertising to grow at the expense of Credit and Finance.

Distribution and retail — NOPE

Much of distribution and retail happen in the physical world, so it’s difficult for me to imagine that AI (which exists digitally and has no physical presence…yet) would cause Advertising to grow at the expense of Distribution and Retail. This really wouldn’t make much sense, as the economy would be directing capital to driving more demand (via Advertising) to fewer places and ways to buy stuff (Distribution and Retail).

Trade — NOPE

As with Distribution and Retail, Trade is very tied to the physical world, and so it’s very difficult to imagine some new trade-off where AI-based Advertising wins share of GDP from Trade.

Mass media, communication, and information dissemination — MAYBE?

This one where there is maybe a connection. There is an argument to be made that the cost of content production by the entertainment industry and publishers is set to plummet because of AI-based efficiencies. Meanwhile, the amount of available human attention will remain essentially fixed. Could the “savings” from content production be absorbed by Advertising? Why exactly would that be? I actually couldn’t say.

One thing to research (maybe such research exists) is the historical share of GDP from entertainment, publishing, and telecommunications over the last 100 years (or whatever’s available). AI isn’t the first technology to revolutionize content production.

Sales and customer service — MAYBE?

This is another one where there is maybe a connection. There are many front-line sales and customer service jobs that will plausibly be automated over the next few years (5? 10? 20?) due to AI. I say “plausibly” because this is far from certain and frankly I hope all us keyboard jockeys still have jobs in 2030.

Given the “last mile” role that these jobs play in converting latent demand into sales transactions, there is a close connection to the role of Advertising.

However, these job roles usually encounter demand that is very close to transacting, unlike advertising which seeks to generate new demand sometimes out of nothing (I present as evidence the invention of bottled water in the 1990s).

Where Zuck Might Turn Out To Be Right (By Cheating A Bit)

Out of our original six, we’ve set aside three components as being generally unrelated for purposes of Zuckerberg’s statement. So let’s assume that in an AI future, there is some logical (if ill-defined) connection between these three remaining components:

1. Advertising

2. Mass media, communication, and information dissemination

3. Sales and customer service

AI means that Meta can essentially take all three under its own roof and incorporate them into its ad platform. Meta can use AI to create the entertainment content that the ads run alongside. Meta can sell ad space and make the ads themselves using AI. And Meta can provide last-mile “agentic” tools to advertisers for sales and customer support.

Meta could bucket all three together as “advertising,” and that group may appear to grow its share of GDP. This, of course, will be because Meta is now aggregating things that used to be discrete, and so the growth in share of GDP is essentially an illusion.

Maybe this is what Zuckerberg meant anyway..

This kind of “vertical integration” has already been the playbook of Amazon and Google to some extent; to package direct-response ads and brand ads in contrived, often counter-intuitive ways. The shift to digital unlocked tremendous new economies of scale and opportunities for vertical integration and it’s reasonable to expect that AI will only magnify this.

Too Early To Say

At this point, things are unconclusive. After pondering this for several days, I can’t think of a clear causal relationship whereby AI causes Advertising to grow at the expense of other components of the economy. The best I could come up with was a cynical scenario where multiple things get conveniently reclassified as “advertising,” but the cynical scenarios are always the ones people miss when pondering the possibilities of miraculous new technologies, so I’ll probably turn out to have been right.

Some ex-Media.Monks colleagues I’ve known and worked with for many years just today launched their new firm Anthrologic, focused on building and deploying AI agents for marketing organizations. Here’s the press release.

I’m really excited to be on board as an advisor. Specializing in data + AI + marketing, Anthrologic inhabits an intellectually fascinating space, but on a commercial basis they’ll do quite well too.

I thought I’d use this news as an excuse to explore an aspect of AI hype that gets overlooked, perhaps because it’s so simple.

It’s easy to poke holes in AI and LLMs, but don’t let this fool you

It’s fair to ask the question, “Are AI agents really a thing or is this just bullshit?”

You can then think through the question from a bottom-up or a top-down basis.

The bottom-up approach has taken up a lot of the discourse, with everyone opining and hair-splitting on LLMs’ reasoning capabilities (or lack thereof), the Game of Thrones between big tech companies, tech architecture, the GPU market, and so forth.

All valid areas to focus on. But when you go into the weeds…wow! A lot of doubts, what-ifs, and buts start to emerge. And so you end up having trouble actually answering the original question: “Are AI agents really a thing or is this just bullshit?”

So let’s take a different approach and look at this from a top-down basis. Instead of starting with AI, let’s begin by looking at digital marketing as a function and discipline.

Marketing is a lot of work—a LOT of work

In digital marketing, adtech, and martech, the difference between the product marketing and the lived reality of marketers (and their agency partners) is VAST. The tech vendors’ product marketing or the visions presented by thought leaders basically always looks like this:

Anyone who’s come remotely close to touching any part of that process knows that it’s incredibly labor-intensive and messy. Each part of that diagram could take months or even years to put into place, even on a rudimentary basis.

To be clear: depending on the size of the company or project, we could be talking thousands, perhaps tens of thousands of person-hours on an in-house and outsourced basis invested to accomplish a FRACTION of the original vision or aspiration. Billions of dollars spent every year around the world to put those boxes and lines into place.

To make matters worse, the marketing industry expends a tremendous amount of energy criticizing itself and introspecting on why —despite all this technology at everyone’s disposal— everything is always such a dumpster fire and why don’t things work like the diagram said they would?

And most importantly: everyday consumers at the receiving end of all this tech and data aren’t particularly impressed with the experiences they get.

Chasing a carrot that humans alone can’t catch

There’s a simple reason that the reality hasn’t matched the aspiration: the digital marketing and advertising industry has consistently (and vastly) under-estimated the amount of labor required to go from concept to reality.

This happens for any number of reasons: poor scoping, ignorance of dependencies, over-promising from vendors, etc. Regardless of the reasons, no amount of tech can really help when the resulting labor always balloons to 10x what you hoped for.

Marketers then look to additional tech to help make up the lost ground. But then every fix and patch, every new piece of middleware always spawns its own 10x burst of scope.

Think of all the three-letter acronyms and trends that have each spawned their own multi-billion dollar industries over the years: CRM, CMS, SEM, marketing automation, SEO, RTB, CDP, ETL, ABM, and so forth. There is no end to this story; no stable equilibrium.

And so actual value always falls short of potential value by an embarrassing amount. Actual labor, cost, and time always exceed anticipated labor, cost and time by an equally embarrassing amount.

The marketing industry has a harsh reality to face: there literally aren’t enough human hours in a day, a week, a quarter, a year to “catch up” and bring reality into alignment with expectations.

You can probably see where I’m going with this…

A new source of labor rides to the rescue

Yes, it’s AI agents. But I’ll belabor the point a bit further:

Anyone who’s tinkered with AI and LLMs can see the potential to hand a lot of rote marketing labor over to armies of specialized bots. Even if the foundational models don’t get any better, we’re already looking at years of new value to capture.

And as I’ve argued, digital marketing and advertising ALREADY suffers from a massive secular labor shortage due to the fact that humans had developed a bunch of adtech and martech that exceeded their capacity to actually USE (it’s a bit absurd, like something out of a Douglas Adams or Joseph Heller novel). The need is there and it’s been there for years even though the industry hasn’t recognized it.

This lack of recognition is partly because we had no counterfactual for comparison. But now we do. AI offers marketers a new way to externalize much of this labor that humans were literally never going to get around to because there was just too much of it.

Is AI a silver bullet? No. Will AI-powered marketing still be imperfect and messy? Definitely. But is there pent-up demand for labor and can AI agents fill at least some of that gap? Yes, I believe they can, and I believe that closing that gap will be a multi-multi-billion dollar market over the next several years.

Anthrologic

Anthrologic was founded by people I’ve worked with for years (namely Tyler Pietz, Adrian Domek, and Alex Gwin). The need for AI can be articulated any number of ways (this essay is just one), but it’s definitely there. In turn, the specialized partner capabilities that will thrive in this new environment are another thing to consider. I won’t get into that right now, other than to say Anthrologic either already has those capabilities or has the proven ability to adapt and develop them as need arises. Hence my excitement at being involved.

If any of this sounds interesting to you and you want an introduction, let me know. Or just drop them a line.

To lots of folks in tech, it feels natural to conclude that AI will disintermediate, replace, and eradicate ad agencies in much the same way tech has disintermediated travel agents, cab companies, newspapers, movie theaters, department stores, stock brokers, and many other businesses.

The thinking goes like this: when AI gets sufficiently good at emulating the work of an agency (an ad creative execution, a campaign plan, a website), then marketers will just “go direct” to AI agents for all their marketing needs and cut agencies out of the picture.

This logic is, however, flawed. It's based on a superficial understanding of (or obliviousness to) at least four things:

1. What marketers do (and don’t do) in real life

2. Why agencies exist

3. Basic accounting & finance

4. The rate at which marketing gets more complex and the phenomena of “induced demand”

How marketers and agencies work together (in the real world)

So what are the respective roles played by marketers and agencies? I mean the roles they play in the real world, not in the idealized fantasies of tech engineers who have never worked on a marketing team.

What marketers really do

A popular premise is that marketers are a hands-on bunch: spending their days buzzing about; brainstorming and launching dazzling campaigns for brand-building, new customer acquisition, customer retention, and so forth.

When you do a Google Image Search for “marketing team,” these are the high-energy depictions you see:

Yes, brand-building, customer acquisition, and retention are some of the functions that marketing serves in a business. But many (I would argue most) marketers focus their efforts on deploying marketing budget and other company resources to external vendors such as agencies, freelancers, consultancies, and tech vendors. It is these external vendors who do much of the actual execution.

(In-house marketing teams are a partial exception to this, but even the in-housiest of in-house teams outsources plenty of labor, technology, and expertise.)

Marketing teams expend much of their effort thusly:

1. Assimilate a set of business problems aligned to marketing.

2. Use the marketing budget to offload those problems to external specialist vendors.

It looks like this:

In short: many of the best CMOs and marketing teams are adept at efficiently externalizing their problems versus internalizing the difficult, time-consuming work and specialized expertise of directly solving those problems.

Keeping time and budget constant (read those words again, then look up the definition of “opportunity cost”), an effective outsourcing strategy will almost always beat the DIY approach to marketing.

(In-housing may seem like an exception to this rule, but the fact is that in-housing requires its own outsourced consulting vendors, careful planning, and airtight strategic justifications; successful in-housers don't just fall ass-backwards into some magical self-serve marketing cockpit.)

Now let’s talk about the role of agencies

Agencies exist to say “Yes” to client asks. In other words: agencies exist to soak up problems clients can’t or won’t solve on their own.

When a client (a marketer) approaches an agency with a problem (“can you fix our SEO?” or “can you draft a media plan to launch this hand cream in Argentina?”), the client has already determined they do not want to or cannot do that work, or cannot accomplish the work themselves in an acceptable timeframe.

Given this scenario, let’s think about everyone’s goals and motivations:

• The client needs an agency to say Yes, because the client needs to externalize the problem and can move on.

• The agency wants to say Yes because Yes means more revenue; saying Yes is how agencies grow.

• The agency can’t say “No” or “we can’t” too many times, otherwise the client will find another agency because clients don’t have the bandwidth to manage a large number of picky or overly specialized agencies.

• The pie is finite, so saying No makes competitors richer. When an agency says No, they are risking that a competing agency will say Yes, win the business, and grow their market share.

Okay, I’ll humor you. Let’s imagine a utopian future.

AI-driven marketing utopians are basically presuming a future where clients “go direct” using AI and cut out intermediary ad agencies that they currently rely on for planning, creative, buying, analysis, and so forth.

Looks good on paper, doesn’t it!

However, it’s complete bullshit. Adtech and martech have been chasing the “self-serve” carrot for decades now and some elevated state of marketing nirvana remains elusive. Every new tech innovation (CRM, CMS, SEM, SEO, RTB, etc) brings with it layers of new complexity and a new wave of intermediaries, gatekeepers, and specialists. Remember when WordPress was supposed to make it easier to build and manage websites?

The Platonic ideal of “self-serve,” disintermediated marketing is an illusion. AI won’t change that.

Reminder: this is not a bad thing! It creates jobs and growth! And new reasons to go on conference boondoggles!

The cost of marketing expertise

In an idealized Post-AI scenario, the marketer would have to hire new FTEs who “know what good looks like” to oversee AI tools in niche areas that agencies used to handle.

Just imagine it: you’re a brave, principled CMO asking your CFO for hundreds of thousands (perhaps millions) in new budget to hire a bunch of expensive specialists who can then be very prescriptive in telling AI applications what to do. Yay! Self-determination.

The first question the CFO will ask is: “Wouldn’t it be cheaper to hire an agency who ‘knows what good looks like’ and uses AI versus taking both the AI and the expertise in-house?”

Well, yes it would be cheaper in fact. Why? Because an agency can amortize the cost of expertise across multiple different paying clients. For an agency, expertise is a client-billable profit center. For clients, in-house marketing expertise is always a cost center (if you don’t know basic accounting principles, please look this up, I beg you).

The same principles often hold true for technology. This is why so many clients access technologies like Demand Side Platforms, ad servers, and audience data through their agencies. AI may be yet another marketing technology that sometimes ends up being cheaper for marketers to access via agencies rather than take in-house.

An AI idealist might push their glasses up their nose and counter thusly: “Increased marketing efficiency driven by AI will free up budget to hire more in-house marketing staff.”

Oh Lord, give me the patience to tolerate such naivete. If AI makes marketing “more efficient” (whatever that means) across the board, then all companies will enjoy the same advantages. Once this happens, each company’s CFO will re-examine marketing budgets and efficiency standards based on new “table stakes” that reflect the company’s peer group.

When a job function or a department becomes more efficient on a secular basis, the CFO doesn’t just give the savings back to that department as a perpetual gift. Those savings are reallocated to wherever they can be most productive in the business.

AI doesn’t fundamentally change the rules of capital allocation and marketing is no exception.

Marketing complexity and “induced demand”

“Induced demand” is the reason that, no matter how many lanes are added to freeways, there’s always traffic. Why? Because the more lanes you add, the more travelers are encouraged to hit the road and the more developers build along those roads.

In other words: the more capacity you add to comfortably meet existing demand, the more that demand will grow, causing demand to always exceed some “comfortable equilibrium” in relation to capacity.

Your hair will always be on fire and everything will always be a complete shitshow.

Sound familiar?

For each thing that AI makes more efficient (or obsolete), the market will invent ten more things. However many “freeway lanes” there are for marketing & advertising, the industry will pile into them like so many Labor Day road trippers.

Induced demand will save agencies.

Ever since the beginning of marketing and advertising as we know them (approx. the last 150 years), the industry has grown more efficient. For example: Adobe Creative Suite enables present-day amateurs to run circles around commercial artists working slowly and diligently with pen and ink 75 years ago. Did Adobe CS kill opportunity? No! Better and more efficient tools like Adobe CS have always helped grow the industry because they expand capacity which then boosts demand.

Increases in efficiency would cause agency opportunity to shrink if industry complexity remained constant; if there were never new media channels, ad formats, consumer habits, economic trends, etc. But this simply isn’t the case. The marketing & advertising industry will continue getting ever more complex and fragmented, and this complexity will cause the demand for marketing labor and expertise to outpace clients’ in-house capacity, which in turn will ensure that the need for agencies will persist.

B-b-but what about super advanced AI?

Maybe our imaginary AI idealist will now say: “Myles, you keep saying that CMOs will need to in-house FTE’s to run the AI. This won’t be necessary because marketers will interact with the AI the same way they interact with an agency today. Marketing teams don’t really need to change! Dude, agencies are gonna die! Admit it!”

Don’t drop that mic so fast. In that hypothetical world, I’m not entirely certain what any of us will do all day whether we work in marketing or not. It’s an unserious argument because when you extrapolate it, you must imagine a business world, an economy, a society, a human race that has been turned completely upside down by AI that is able to flawlessly emulate human enterprise. Maybe this will happen, maybe it won’t, but I don’t see how present-day ad industry commentators would be able to predict this any more accurately than a cave person sparking a fire could predict the steam engine.

If we’re going to stretch that far into the future, why stop at AGI? Why not predict the death of marketing itself when all of human consciousness has been uploaded to spacefaring supercomputers?

Time to summarize

Let’s recap:

• Marketers don’t necessarily want to in-house the work and expertise of agencies. This is why they hire agencies in the first place.

• Staff, labor, and expertise are expensive. All things being equal, CFOs don’t like hiring FTEs or in-housing technology for work that can be outsourced at a lower cost. AI doesn’t change this.

• Marketing and advertising will continue growing more complex. As the goalposts keep moving, marketers will always be in need of agencies to make up the difference.

• As AI changes how marketing work gets done, agencies will adapt as they always have.

The last point I’ll raise is that some present-day agencies may indeed get killed by AI. I’m not trying to convince you that every individual agency is impervious to AI disruption. What I am trying to demonstrate is that AI isn’t likely to cause the agency business to die in aggregate. I simply don’t see how AI fundamentally changes marketers’ aggregate calculus for what they take in-house versus what they outsource, nor do I see the sheer volume of work really changing as efficiency gains from AI are offset by an increasingly complex marketing landscape.

Cheers for now! Thanks for reading.

This post is just a thought experiment. It’s not a prediction and it’s certainly not based on any direct knowledge of Meta’s strategy. This post was inspired by two facts:

1. Meta still stands out as a major “big consumer tech/media/commerce platform” that DOESN’T offer cloud computing.* The three platforms of this kind that do offer cloud computing are Amazon (AWS), Google (GCP), and Microsoft (Azure).

2. Meta (like many other companies right now) is taking a big gamble on hoarding expensive AI computing power that will need to be monetized ASAP.

* Apple would be another example, but that’s a post for another time.

Meta’s big bet

In case you haven’t heard, Meta is making an enormous bet on AI and AI infrastructure. Here’s what Mark Zuckerberg said on 18 Jan 2024:

“…we're building massive compute infrastructure to support our future [AI] roadmap, including 350k H100s by the end of this year -- and overall almost 600k H100s equivalents of compute if you include other GPUs.”

(source: Instagram)

Raymond James estimates that these hundreds of thousands of specialized GPU (Graphics Processing Unit) chips are going to cost Meta close to $9 billion.

This hoard of GPUs is a big part of a big bet. Overall, Meta’s R&D expenses (as a percent of revenue) are currently roughly double that of Amazon, Google, and Microsoft:

Q1 2024 R&D expenses:

• Amazon: 14%

• Google / Alphabet: 15%

• Meta: 27%

• Microsoft: 12%

For historical comparison, Meta’s Q1 2021 R&D expense was 20% of revenue. While the leap from 20% to 27% in 36 months doesn’t strike me as constituting a risk-it-all existential moonshot, the magnitude of investment has been significant. And the AI arms race —with its voracious appetite for compute capacity and richly compensated technical staff— seems to be shaping up to be a giant money pit for tech competitors. For example, Zuckerberg recently cited the need for sheer electrical power as being yet another front opening up in the AI wars.

An emerging war of attrition among tech giants

Meta isn’t the only company digging an AI moat and filling it with GPUs (and maybe nuclear power plants). If we assume that we’re still in the early skirmishes of an AI battle royale, Meta is signing up for a long, costly war.

How long can Meta take to figure out how to monetize its GPUs?

Meta investors want assurance that R&D expenses (whether related to AI or not) will be levered into profits on some reasonable timeline. However, the “killer apps” of AI are still ambiguous and AI tech itself is far from fully baked. Tech giants like Amazon, Google, Meta, Microsoft, and OpenAI are probably still years away from figuring out how to actually make money from AI.

Meta needs a lot of things to go right in order to traverse the chasm from present-day R&D outlays to future payback.

Do I personally believe big bets on AI are justified? Absolutely, but there’s no getting around that bets are bets; they require risk and cost.

If you strip it down, Meta is betting that they can find at least one breakout AI “killer app” (whether internal, user-facing, or B2B) that starts to pay back the billions of dollars they’re investing before they have to start writing those investments off as losses (and possibly face more existential threats from competitors who were actually successful in finding AI killer apps).

Incremental AI-based improvements to Meta’s ad algorithms probably won’t cut it

Currently Meta makes around 97% of its revenue selling ads. Juicing this with AI is an obvious answer for how Meta makes its Big Bet pay off.

Meta is almost certain to improve its ad platforms and ad monetization using AI. AI could build better target audiences, AI could craft more effective ads, and AI could even create landing pages and what-not.

But is this really a “killer app?” Or is it just another incremental improvement in digital advertising’s long history of applying machine learning and AI to the problem of selling the most relevant consumer eyeballs to the highest bidders? In other words: would a better ads algorithm represent a step change or paradigm shift? Not really. Would it materially grow Meta’s TAM or unlock any new revenue streams (either on a consumer or B2B basis)? Not really. When a car company adds 50HP to next year’s model, they don’t expect to turn the car industry upside down.

Better ad algorithms using the latest advancements in AI are the new table stakes. Amazon will have them, Google will have them, everyone will have a better black box. There’s no reason to assume it would confer any special profit advantage to Meta and therefore little reason to assume that “making a better ads black box” is the reason Zuckerberg is sticking his neck out right now.

Could Meta rent out some of that compute power?

Instead of betting all its chips on its own ad business, could Meta launch a cloud platform? Such a platform could essentially “rent out” coveted GPU capacity to global businesses and startups.

Directly monetizing at least some of their GPU and specialized AI infrastructure is something that Amazon, Google, and Microsoft are already able to do via their own respective cloud platforms.

I couldn’t say what’s involved in establishing a cloud computing platform. I assume the investment levels and timelines are significant.

However, AI is new and GPUs are specialized. Meta could be nimble. They wouldn’t have to build a comprehensive general-purpose cloud behemoth like AWS with its dizzying array of over 200 services. Meta could offer a handful of specialized and differentiated AI-based cloud services that it could bring to market faster and at lower cost.

Setting aside the effort to stand up a cloud computing platform of any kind (glib as that may be), it seems that such a platform would shorten the “journey” that revenue would have to take in order to monetize GPU capacity. An improved ads algorithm is a more indirect way of monetizing GPU capacity than simply renting out the “raw” capacity via a cloud computing platform.

Lastly, it’s worth noting that improving its ad algorithms and launching a cloud computing platform are not mutually exclusive monetization options for Meta.

Is there an opening for a new kind of cloud platform?

This is where the thought experiment gets really interesting.

Let’s be imaginative and assume that AI will unleash new computing paradigms. Perhaps ones where computers and systems aren’t “programmed” or configured any more.

Instead of the burden being on the human user to carefully input instructions into narrowly-defined applications, perhaps AI-based systems will be more like responsive, versatile, general-purpose organisms that will obviate the need for step-by-step instructions and generate their own path to achieve whatever goal the user provides.

We can already see this with generative AI: instead of a human using a cursor to meticulously draw a cat in a highly specialized piece of drawing software, the human just says “draw me a cat” and generative AI figures out the steps for itself and draws the cat.

Just think about how different this is! There’s no cursor, no specialized drawing software, and no individual brush or pen strokes. It’s faster, it makes (mediocre) cat drawings even more abundant than they already are, it eliminates the need for specialized artistic skills, and it consumes orders of magnitude more compute resources.

We’re about to experience a radical change in how computers work and what differentiates effective computer applications (and infrastructure) from ineffective ones. I couldn’t say exactly what form this will take, but I would be willing to bet that Mark Zuckerberg agrees with that basic premise.

One of the most elegant illustrations of this revolution was tweeted by Dharmesh Shah, co-founder and CTO of Hubspot (who happens to be on Substack):

Now think about cloud computing, which has been around since 2006 when AWS launched its first services. That’s 18 years. All the current cloud platforms were designed from the ground up to break the shackles of the prior computing paradigm and disconnect computing power and storage capacity from the constraints of local machines and on-prem (or colo) infrastructure.

Revolutionary for its time, but not necessarily fit for purpose in an AI era.

If you believe AI is going to cause a tectonic shift in computing, then it stands to reason that cloud platforms carrying the baggage of the prior paradigm might be at a disadvantage when faced with a new competitor designed from the ground up for the new paradigm.

So in addition to monetizing their GPU hoard more directly, might there also be an opening for Meta to drive a wedge into the cloud computing market with a new platform designed exclusively for the AI era?

What would a cloud computing platform look like in the AI era?

Before I go further: yes, I know that AWS, Azure, and GCP (and all the rest) offer AI services, access to GPU capacity, and all that good stuff. I’m not predicting these platforms are going to die. What I’m suggesting is that like any incumbent company, the assets and innovations that lifted them to dominance may become a drag on their ability to pivot during a time of disruption. This has happened time and time again across many industries and there’s every reason to believe that today’s cloud computing giants may struggle to adapt to an AI era that totally redefines which kinds of infrastructure are valuable / differentiated and which are not.

So if we’ve drawn a line between “old” and “new,” what is it that might define “new?” If Meta were to launch a cloud computing platform, what might differentiate it in substantial, “paradigm-ey” ways?

1. Focus on application outputs, versus typical cloud platform obsession with meticulous user inputs and byzantine configurations (if you’ve ever used a cloud platform, you know of which I speak).

2. Native integration into Meta end users. Cloud platforms are mostly built on a “bring your own data” model (as well as a “find your own customers” model). This is a tough pill to swallow in the AI era because only a handful of companies in the entire world have enough data to train good standalone AI applications. Amazon, Google, and Microsoft, for example, all keep their end users and end user data (from services like Amazon Prime, Google Search, Gmail, or LinkedIn) completely firewalled from their cloud platforms. Might application developers be attracted to a Meta cloud platform that comes with built-in training data and a built-in global user base of billions?

3. Pricing. AI may rapidly erode the value of “bread and butter” cloud computing services that are usually consumed on an unbundled basis (even simple cloud applications can require dozens of discrete cloud services, all separately racking up consumption bills). Builders may come to expect extreme levels of abstraction and bundling due to AI’s ability to “eat” other software. This may affect pricing. I couldn’t say exactly how, but the incumbents will have trouble adapting to it, as it will mean killing their current flock of golden geese.

Those are just a few ideas. I expect someone somewhere has done a much deeper dive. I also expect the incumbents are already thinking about this and planning for it.

The “thousand monkeys” strategy

The last point I’ll make is that Meta is currently a single company looking for AI killer apps. If Meta were to launch a cloud computing platform, they would be incubating potentially millions of builders around the world. A cloud platform could be a way of “crowdsourcing” the discovery of AI’s killer apps and making sure they come to life in the Meta ecosystem versus Amazon’s, Apple’s, Google’s, or Microsoft’s (among others).

That’s all I got

If you got this far, thanks for reading. Are there other arguments in favor of this that I didn’t think of? What are the arguments against? What would an “AI-native” cloud computing platform look like?

This strikes me as wildly non-intuitive on a macro basis. Are RMN’s some sort of Perpetual Motion Machine for advertising? Will the RMN pie just grow and grow as more ads are jammed in front of shoppers by more retailers and me-too RMN’s?

Retail media draws its value from brand spend

Let’s think for a moment how shopper demand is generated on a macro basis. Retail media exists "down-funnel" from TV and other mass media channels:

1. Mass media like TV (whether linear or digital) drives brand awareness and shopper desire.

2. A desire to buy things drives shoppers to retailers.

3. At this point the retailer can monetize shopper traffic via an RMN.

Those are the dominos that have to tip over in order for RMN's to work. Note that the first domino has nothing to do with the RMN.

Brands are investing in two places: first to spark demand up-funnel (in mass media) and second to put their thumb on the scale and win the sale at the shopper’s point of decision (via RMN’s). An RMN’s value isn’t generated or sustained solely by the retailer and retailers can’t just take shopper demand for granted.

How much is a sponsored listing worth in a world with no shoppers?

I do understand that a retailer has a loyal, built-in customer base and valuable retail infrastructure irrespective of brand advertiser spend. However, if all non-RMN advertising suddenly went dark, retailers would start to see declining traffic and revenue as shoppers lost interest in buying stuff or trying new products.

Shopping is a byproduct of advertising, not the other way around.

And if shopping is a byproduct of advertising, there must be an inherent cap on the total size of the RMN market.

It’s physics, people

Put another way: some of the logic around RMN's is starting to sound like sitting in a sailboat and blowing on the sail and expecting the boat to move.

Things just don’t work that way. The wind has to come from outside the sailboat and the speed of the sailboat is limited by how much wind there is. Nor can a sailboat simply add more sails; more sail area doesn’t let you “outrun” the wind.

(It is at this point in the analogy that I can envision some cheeseball vendor claiming they will provide speedboats mounted with giant fans to make the sailboats go faster…and you really can’t blame the vendor because you know a lot of buyers would fall for it)

As more retailers pile in and launch RMN's (or add inventory to existing RMN's), the aggregate size of RMN's (in terms of total available ad/sponsor inventory) will exceed the demand generated by higher-funnel brand spend and the ecosystem will have to rebalance. Beyond a certain point, you’re only going to sell so many boxes of Cheerios no matter how many sponsored listings you buy.

Perhaps not EVERYTHING can be an ad network.

Pardon me while I guffaw and slap my knee. You don’t “try” advertising. B2C companies advertise and engage in other paid promotions because the competitive environment demands it. Follow the logic all the way upstream to its source and it is the combination of capitalism and free speech that inevitably leads companies to advertise.

This inevitability takes different paths for different products, companies, and industries, but every consumer goods/services company (provided they last long enough) eventually has to pay money to get and keep the attention of consumers. This can be counter-intuitive to the kneejerk “advertising is for losers” crowd because advertising actually gets MORE inevitable the more successful and entrenched your company becomes. Paying for consumer attention isn’t a mark of failure, it is a mark of success and a Good Problem to Have.

Over a year ago, I posted a thread to Twitter that games out a few paths that I felt would eventually lead to Tesla adopting paid advertising and promotion. Most of what I said has held up, and so with yesterday’s comment from Elon Musk, I figured I would share it.

I’ll link to the first tweet (from 18 Feb 2022) and post the entire thread below. Twitter doesn’t let Substack embed tweets any more [sad face].

https://twitter.com/Myles_Younger/status/1494720300704960514?s=20

1/ Tesla = interesting case if you believe in advertising. The “we don’t need to run ads” cases are the most compelling. I *def* believe in advertising, & so I love to think about the case of Tesla, esp now that they are clearly a success story & EVs overall are here to stay.

2/ Def worth pointing out that Tesla has had the distinct advantage of competing against incumbents who, by nature of their business, cannot pivot quickly.

3/ The “make a genuinely desirable EV” category was *wide* open. Priuses are dorky. I can’t actually name another hybrid/EV model. However, exploiting this required massively risky upfront expenditure…it wasn’t like there was a red carpet for Tesla.

4/ And this is where it helps to have an Elon Musk: unequivocal vision and, especially in hindsight, clearly able to start a car company. Has anyone in the US in the last ~100 yrs successfully launched a brand-new mass-market car company?

5/ But here we run into two broader problems:

1. Any successful company will eventually have competitors and see its differentiation commodified away (srsly what’s the diff btw a Ford and a Chevy?)

2. “Celebrity” CEOs don’t scale. You can’t manufacture them.

6/ And *not* going bankrupt as a global auto manufacturer (R&D expense and fixed overhead is *incredibly* high) requires PREDICTABILITY. Honda can’t align on a ten-year global EV plan based on “finding their own Elon Musk.” Really dumb idea doomed to fail.

7/ Tesla was/is fortunate to have a CEO like Musk. But it’s just dumb luck. Elon Musk isn’t something you can whip out your checkbook and buy off the open market. You either hit the jackpot as a company and get a Musk or a Jobs or you don’t. Nearly all companies don’t.

8/ Musk is able to *personally* influence public perception two main ways that I can see:

A.) He’s laser focused on making EVs that people actually want. Like *really* want because they look good, have hard-to-find features (eg battery life), and aren’t wimpy.

9/ Side note: yes I know Teslas are flawed. But you can’t tell me people haven’t been falling all over themselves to buy them. People *want* them and that’s not an accident, probably largely due to Musk.

10/ B.) Musk himself is a well-known and notorious public figure. He’s a carnival barker who can make his own news and be a living, talking, controversy-making embodiment of the Tesla brand. By contrast, can you name a single Toyota executive?

11/ Let’s look at the “what about advertising” question from the POV of Tesla’s competitors: So how do these poor wittle multinationals stuck without their own Elon Musk shape public perception? That’s right: with advertising!

12/ The other thing to think about is that when it comes to the products themselves (the cars), Tesla’s competitors hold all the cards right now: in the next ~2 years (number out of my ass) Tesla’s competitors will commodify nearly all of what made Tesla originally unique.

13/ And what is Musk’s “share of voice” in the EV market versus a major autos OEM? Maybe at one time Musk’s was the dominant voice in a tiny market, but a few quarters of major TV ad campaigns will make him seem like a sideshow.

14/ And then what about Musk’s power of celebrity? His personal brand equity is worth untold hundreds of billions. It’s formidable. He has a “moat” in that his competitors can’t just mint their own Musks on demand.

15/ So what’s a brand to do??? HIRE CELEBRITY ENDORSERS. That’s how brands manufacture from whole cloth their own notoriety and celebrity halo effects. And they channel this through paid advertising (and a bit of PR, like when the POTUS drives your EV around the White House).

16/ So we have established that from the POV of Tesla’s competitors, Tesla does not have a monopoly on:

• unique products

• the power of celebrity So what are Tesla’s options?

17/ It could use “product-led marketing” and redouble efforts on “out-innovating” the major OEMs. But this can’t last forever. That’s committing to fighting a perpetual multi-front war against a diverse group of *very* well capitalized competitors.

18/ And keep in mind that there are many forms of innovation that a consumer doesn’t directly experience by way of the car itself. Logistics and supply chain innovations, manufacturing innovations, consumer financing innovations, distribution innovations, etc.

19/ Tesla could also commit to Musk’s personality itself. Both to continue to own the product roadmap and to go “direct to consumer” via channels like Twitter and earned media / PR coverage / tabloid coverage. Other OEMs *literally* cannot buy this kind of marketing.

20/ But when does Tesla hit a ceiling on how many global EV buyers give a rat’s ass about Elon Musk? Given the utter scale of the shift from gas-powered to EVs, probably sooner that Tesla would like! The market for EVs is probably close to a billion people worldwide!

21/ And with other OEMs flooding the airwaves and setting public perception on what comprises a “cool” or a “normal” EV, Musk’s personal vision of Tesla’s roadmap gets less and less relevant because he’s no longer owning the conversation in a vacuum.

22/ Actually gets increasingly risky as time goes on to place the entire “weight” of Tesla on one man’s shoulders.

23/ So from Tesla’s side, it doesn’t seem wise to rely on Elon Musk personally to:

• own the product roadmap (or at least set that perception; I realize Tesla employs more people than just Musk)

• embody the brand

24/ Sooner than many expect, I think Tesla will have no choice but to evolve beyond its early strategy and invest heavily in paid advertising.

25/ Think of Tesla as a case study in why “nearly all roads lead to advertising.” Hate to say it, but there really aren’t any “cheat codes” around this, even in 2022. Brands that run ads aren’t stupid or inferior. Ads are just what happens when you game it out.

If you made it this far, thanks for reading. Viva advertising.

For a few weeks, I have been drafting a Substack post explaining why consumer behavior is about to radically change because of AI. Finding time to finish the post has been tough.

But this afternoon I posted a tweet thread that serves as a “preview” of that post. I am embedding that thread here.

My thread was in reaction to a tweet by Harley Finkelstein, President of Shopify:

The future of how people buy

This was my reaction:

Advertising will never be the same again (again)

That was it for my short tweet thread, but I’ll add the following:

My point (or prediction) is not that AI will cause advertising to cease to exist, but that advertising will change. This change will likely be quite painful.

Billions and billions of dollars of ad revenue and media planning are currently predicated on a particular set of consumer behaviors that only date back about 25 years: people navigating around websites and apps; reading, typing, and clicking. In fact, we do so much fussing over sites, apps, and devices that the average American is living a 32-hour day when you account for multitasking.

Until now, consumers’ frenetic behavior in the digital sphere has been a great boon to the media and ad world, as it has meant an ever-growing pool of valuable consumer attention to be monetized.

However, consumer behavior is about to change, and so too will advertising.

More to come on this in a future post.

Subscribe now

Filed under: AI and consumer behavior

Web browsers are a really old technology.

It’s 2023 and web browsers are around 30 years old as a technology. This age is apparent when you look at the level of manual effort required to use one. It takes an awful lot of typing and moving cursors around and clicking in order to get a browser to do anything useful.

We have even built new features allowing us to invest more manual effort into browsers. Want to multi-task beyond the limits of human ability? You can do that! Just open a hundred new browser tabs on top of the hundred tabs you already have open!

When web browsers debuted in the mid 90s, there wasn’t as much information on the web for people to process, and all of that information was in static form. But now, in 2023, none of us can get anything accomplished online without manually flipping back and forth between dozens and dozens of browser tabs, logins, and web applications (some of which may be synched to entirely separate devices).

Cookies are a really old technology.

Browser cookies are also close to 30 years old. And why do they exist? To aid we humans in all this manual cross-referencing we’re always doing inside web browsers!

Think about it: a browser cookie doesn’t move the ball forward for you. If I’m logged into Hotels.com, all a cookie does is maintain my login state and remember where I left off. A cookie simply reduces the effort required to get me, the human, back to where I can do more heavy mental lifting: searching for travel dates, comparing rates and properties, cross-referencing with airline sites, etc.

In this way web publishers are more like supermarkets than restaurants. A web browser helps you shop the aisles for ingredients, but it doesn’t get you what you ultimately want, which is an amazing meal.

ChatGPT is showing us a new way of interacting with the web.

Forget using prompts to write little essays. That’s really, truly impressive stuff, but why are we really all so amazed? Why has ChatGPT captured people’s imaginations?

For once we’re being served an amazing meal online without having to shop for all the ingredients and do the cooking ourselves. We’ve never actually seen the internet do this before and we’re delighted.

If a chef is making my meal, why do I need a pencil and paper for a shopping list?

The advent of AI-based interaction will transform what a web browser is. How many of your daily points, clicks, and keystrokes could be eliminated by AI? If ChatGPT is any indication, I would venture north of 99%. The next logical question is: why would you still need an app you basically no longer interact with?

I have no idea exactly what will replace browsers, but I don’t see the web browser of today surviving into the next decade. Or maybe its “survival” means moving into secular decline with an aging and shrinking user base of die-hards who insist on doing things the hard way (think cable TV).

The death of the cookie may not matter very much in the end.

One thing very few have bothered exploring in the last few years as third-party cookies’ final demise looms: what if browsers themselves give way to a different and better technology? Why assume browsers themselves will remain perpetually relevant? It feels like we’re in the 1980s debating how inkless typewriters might work. It doesn’t matter!

And remember: the three most popular web browsers are made by Apple, Google, and Microsoft. These also happen to be three companies which will surely be launching all manner of AI-powered apps in the coming years (more on that in another post). How convenient for them!

Apple, Google, and Microsoft can deliberately sunset browser technology at a time and in a way that’s advantageous to them. So don’t think these companies aren’t in control. As the rules of the game change, they can simply change the game; this is a key truth that’s being overlooked.

Google, for example, may not fully control the death of third-party cookies, and many aspects of its business will be disrupted, but Google is in a great position to stage manage the decline of Chrome alongside the concurrent emergence of new forms of interaction that are higher-value and stickier.

The advent of ChatGPT shows us that browsers need Google more than Google needs browsers.