My partner was fortunate to have 12 weeks of maternity leave and I, as an independent consultant, was able to modulate my work to be able to spend a lot more time with Nelson in his first few months than I was able to with my first child, Marion (who will be 11 in just a few days), when I was working in the “pre-woke” corporate world with only 2 weeks of parental leave. I didn’t realize how much I would appreciate this time.

Nelson’s pediatrician has informed us that he is colloquially known as a “happy spitter”. Nelson is already demonstrating this with some of the most remarkable comedic timing I’ve ever seen an infant deploy:

Based on prenatal in utero measurements, Nelson was estimated to be a nearly 10 pound baby at birth and, as mentioned, he came into the world as a fairly average 8 lb. 3 oz. I’m happy to report, however, that during his 4 month checkup he weighed in at 16 lb. 6 oz. This is a remarkable trend, doubling his weight in just four months! If this rate of doubling continues, Nelson will be a 4,000 lb. by his third birthday. See plot:

The data is the data, folks.

Mom has been doing great, by the way. I’m not surprised, but always impressed by her patience and capacity to do all the things only she can do and things that I can do but she does better.

Speaking of data, I’d like to ask you as subscribers to this budding young newsletterblogcommentary, which half-finished articles would you like me to finish next?

In the meantime, Nelson and I will be working on the NYT Crossword Calendar we got for Christmas. So far, we’re only a few days behind.

Appreciate y’all, and thank you for the patience while I bonded with my new dude.

Share Artificial Economics

As far as I know, it’s one of the first works on consulting best practices, one so many consultants have been riffing, remixing, and basing our own frameworks upon in the nearly two centuries that have followed.1

Perhaps you recall the bones of the story, but I encourage you to read or listen to it again as a refresher. It is now in the public domain, so it is freely available to read or listen to on websites such as this one. I listened to it on Audible and its free if you have any kind of plan.

But in case you’re not gonna go do that right now, here’s my own rough paraphrase:

There’s an emperor who likes to dress up in fancy outfits, get paraded around town, and receive praise from the citizens.

One day, a couple consultants from out of town tell him they’ve seen fancier outfits with a lot of their other clients.

They tell him they can make him one if the emperor can pony up a couple spools of gold thread in advance.

The guys also tell him that the clothes are so fancy that they’d be invisible to anyone falling behind peers in their roles and stuck on “legacy thinking”.

The emperor was seeing the bigger picture. It wasn’t just an outfit, it was tool; An opportunity to gain advantage over adversaries and allies alike.

He gives them the gold thread and tells them to begin the work. But the emperor is no fool, of course.

He has his Chief of Staff and his Chief Fashion Advisor check in on the work, each independently.

After a few days, the emperor’s Chief of Staff checks in and see the consultants fiddling with the spools of gold, but doesn’t see the outfit.

Suspicious, but unfamiliar with the technology and fearful that he may be stuck in legacy thinking, he gives a positive report back to the Emperor. “Everything is on track for end of week delivery, dear Emperor!”

Two more days pass and the Chief Fashion Advisor drops in on the consultants. Noticing the spools of gold thread looking fairly depleted, he asks if the outfit is nearly complete. “No issues,” the consultants reply.

The advisor barely admitted to himself that he couldn’t see the outfit, and didn’t dare admit it to anyone else. He suspected he was falling behind his peers in other empires, but now that belief had been validated by outside consultants!

Without his own point of view the Chief Fashion Advisor borrow’s the consultants’ confidence. “This innovative outfit will truly reinvigorate your wardrobe, unquestionably!”

Finally, the weekend rolled around and the emperor went to pick up his new threads from the consultants’ office.

Strangely, the emperor saw no gold thread and no outfit, but the consultants were behaving as if they were presenting something to him and his advisors looked on assuringly.

“Here you are emperor! The finest and most innovative outfit in all the lands! Of course, you’ll notice that despite its extravagance, the golden fabric is so light its weigh is imperceivable on the wearer. Shall we help you put it on?”

“Um, yes…assuming my advisors still agree with their original quality assessments now that the product is finished?” asked the emperor.

“Oh, yes!” said his Chief Fashion Advisor without hesitation. “I concur most completely!” his Chief of Staff added.

And so reluctantly, the emperor, unwilling to admit that it was he that had succumbed to legacy thinking and that he had fallen behind his peers in neighboring empires, donned his new clothes.

“WOW!” said the consultants in unison. “Truly and obviously the best-dressed emperor in all the lands!” said the Chief of Staff. “This apparel will incite a cultural revolution! Even bigger than the necktie!” said the Chief Fashion advisor.

And so the emperor, hoisted up on his mobile throne, to be carried through town on his regular route once again, feigning confidence until he could receive reinforcements from his adoring citizens.

Rumors had swirled for days about the consultant-ware’s strange invisibility property, hiding it from view of those not worthy of their station in society.

As the emperor and his entourage arrived in town, the crowds of citizens looked up with confused faces until his advisors provided a prompting.

“See now upon our Dear Emperor the most magnificent fashion statement in the empire and beyond and offer your undying adulation!”

And so all the citizens cheered and shouted praise at the emperor, none of them wanting to appear unworthy amongst the other townspeople.

All the citizens except for one: a child no more than 10 years old.

The child, alone in the crowd of very adult and important citizens, advisors, and the emperor himself, cried out, “Don’t you see? Can you not see?”

The crowd, growing quiet, looked at the child. “The emperor is wearing no clothes!” said the child, laughing a child-like laugh, not mean, just observing something silly.

The crowd whispered and soon the whispers erupted into a chorus of laughs.

The emperor’s advisors whispered as well, a stone gaze of the emperor upon them. But neither they nor the emperor seemed to be in a laughing mood.

The emperor, still upon his hoisted throne, demanded to be returned to the consultants’ office immediately. And so he was, only to find the office vacated and his spools of golden thread gone.

Laying on a table, the emperor saw a book with the title Launching Your Wardrobe Consulting Business: No Capital, No Expertise, No Problem!

The consultants are currently working on a podcast series sharing their framework for success to a growing like-minded and ethically-fluid audience from the next empire over.2

Definitely not a perfect retelling nor the summary I intended, but seriously, the themes in “The Emperor’s New Clothes” kept recurring so much in my professional life as if it were calling out from my childhood memories and demanding re-examination. After doing so, I found a lot more to unpack than I expected. The story brings to the surface a cognitive dissonance at the intersection of my personal and professional identities that has evaded resolution for my entire career, and yet always felt like my own competitive advantage.

I’ve found one of the more challenging aspects of managing my neurodivergence in the modern workplace is seeing the stark relief of the utterly obvious against a background of customary professional etiquette, overly massaged messaging, and general conflict aversion, particularly in group meetings, and finding a way to check my impulses in favor of my longer-term ambitions. Yet, on balance, I’ve found that I create the most value when I can be the child. And with some irony, now as a consultant, I can be the child for my clients.

And in the era of AI, they need the child more than ever. Not because AI is inherently bad, but because when change is so rapid and the stakes keep getting higher, you need someone to tell you the truth. AI assistants are wonderful tools for many things, but they tell you what you want to know. That’s not always the same as the truth.

This was the first in a multi-part series that I’ll be revisiting from time to time, exploring the themes of “The Emperor’s New Clothes” from the perspectives of the various characters and the lessons to be learned about consulting, business, relationships, and human nature that are served up in abundance in a little over 1700 words (in an English translation).

Share

Subscribe now

Ok, fine. So what? What does Stack Overflow’s decline have to do with artists who are being displaced by AI image generators or musicians being put out of work by Suno or Udio? Why lump creators in with technical domain experts?

In the AI Economy, We Are But Data Sources

From an AI model builder’s perspective, if I want to build a system that can replicate these creative styles, I need a lot of training data, i.e. artists’ works. Similarly, if I want to build a system that can write computer programs, I need a lot of training data, i.e. programmers’ code. If I build these kinds of systems, I can dramatically lower the price and speed of access to each kind of expertise, disrupting these existing markets and attracting customers and capital alike.

What is the source of training data for AI systems? Technical domain experts create insight from data. Creatives inspire insight through their stylistic choices. More that just raw data is needed to train AI models. Insight is needed. Context is needed. As AI labs pursue “AGI1”, they need not just the raw data, but the synthesis of that data through the application of expertise in context.

Looking again definition from last time:

Domain expertise is specialized knowledge, creative abilities, and practical skills that traditionally commanded premium prices due to their scarcity and the significant time, talent, or experience required to develop them.

The work product of some domain experts, like software engineers, exists in the digital domain by default. Others, like musicians, visual artists, or writers, have moved to the digital domain because their audiences have moved there. In both cases, most publically available digital work products have been incorporated into AI models already.

Disruption in the Expertise Supply Chain

AI innovations have created disruptive differentiation by offering access to many kinds of expertise faster and cheaper than existing solutions (i.e. human experts, hours of “doing your own research”, etc). Demand is increasing for domain expertise (see Jevon’s paradox) even as economic price plummets. This may also feel like a surge in supply (oversupply!), but this is the counterintuitive piece: the supply of domain expertise is being reduced2.

Why? Back to Econ 101: when prices fall, the amount of supply in the market decreases to the level the market price can support. The rapid rise in the capability of AI models has increased the elasticity of previously inelastic kinds of expertise.

Perhaps you’ve noticed the job market for software engineers has been pretty terrible over the last 18-24 months. For software engineers who entered the job market post early 2000s dotcom bubble burst, the last 2 years may have been the worst since the subprime mortgage crisis of 2008.

But the current AI era is just the most recent in a long, sordid history of disruption to creative fields. We don’t have to look too far back for a notorious example.

The Napster Disaster

In the early days of internet file sharing and MP3-encoding, the file sharing site napster.com was the most well-known place on the internet for finding and downloading music MP3s for free. As someone with disposable income inversely proportional to the size of my rock & roll dreams at the time, I didn’t have much sympathy for James Hetfield and the Metallica boys who were calling this out as piracy. After all, people were listening to their music, right?

Well…I’ll forgive myself for being a bit naive and missing a strong market signal that I would find becoming a professional musician very difficult.

Here’s question I often find myself asking others when discussing the AI economy because the answer startled me when I became aware of the answer.

When did the US music industry finally reach record revenue again after the decline during the Napster era?

The answer? Depends a bit on how you count, but according to the Recording Industry Association of America’s data the answer is not until 2021.

Declines began in 2000, bottomed out in 2014, and have been rising on streaming revenue since. Note the mid ‘00s to mid ‘10s era. The kludgy and short-lived paid digital download model gave some relief, but not until recently has the industry actually returned to real dollar growth. Adjusted for inflation, it’s still only at about 65% of where it peaked in 2000.

After a brutal 2 decades and tenuous, market-forced relationships with new tech partners, the music industry will need to prepare for the next wave of disruption. It seems it is finally starting to wake up a bit with a few lawsuits filed by large catalog owners prompting some closed door licensing negotiations.

In recent news, a new competitor to Suno and Udio has entered the market. ElevenLabs AI Music Generator partnered with 2 large independent music catalog owners (Merlin Network and Kobalt Music Group) to build their model from properly licensed material. By compensating the domain experts who provided the training data for their model, they’ve not only demonstrated a more ethical path than the “ask for forgiveness and a licensing deal (later), not permission” approach that most commercial model builders have employed. They’ve also shown there is an emerging market for this catalog of music/data, which should encourage others to participate. As model vendors compete on quality, the need to better curate that data is also emerging.

How Digitized is Your Expertise?

Things like text, music, video, images, and code are the things we’ve seen front and center in the rapid digitization of the world, but most domain expertise is not so readily captured and visible. Many domain experts have deep knowledge of a narrow space that is not well captured in digital form today, at least not in a way that can easily be used to train AI models. Even if your main work product is one of those listed above, it’s unlikely that the context you created that work within was well captured. Considering what’s at stake in the AI economy, however, you should assume that someone is trying to capture that knowledge and decide how to respond, no matter what your area of expertise.

Subscribe now

Context is always changing.

This means that no matter what AI models have already trained upon, they will always lag behind in access to context. They will continually need new data that exists within present contexts to maintain their performance on expert-level tasks.

Analogously, the music business is no longer facing the existential threat of digital music downloads. The context has changed. The barrier of access to music has been reduced such that the inconvenience of downloading music, even if it were free, is no longer seen as desirable to most consumers. New threats are emerging in the new context, but we’ll save that for a deeper dive in another article.

Bottom Line: Time to Start Paddling

Both technical domain experts and creatives whose work outputs are often publicly available in digital form on the internet are being treated the same way by most AI labs (i.e. not being compensated for their work), despite being a critical part of the AI model supply chain.

Domain knowledge, expertise, and creative style are the raw materials that AI labs process to deliver their end products.

The question for all domain experts becomes “How do I negotiate a better price than “free” for the raw materials I have to offer?”

For experts, there are many strategies to consider, but they all boil down to one choice: Ensure your expertise remains both scarce and relevant, such that you can charge a premium for it; or curate the training data set yourself and find the right buyers. Making the right choice depends on many factors, some of them related to the nature of your expertise, and others related to the state of AI tech. I’ll dive into this choice in future articles. That’s all for now.

Defining “Domain Expertise”

It’s been convenient for me to put out a few pieces and throw around this term as if it’s well understood, but I’d like to get more precise about this terminology to offer readers more clarity. I’ve been using quite a few terms synonymously:

• Domain expertise

• Domain knowledge

• IP (intellectual property)

• Creative style

• Know-how

• Subject matter expertise

These are not new ideas and have been fairly well-defined and understood for a long time, but they are not synonymous. Creative style is very different from subject matter expertise in nuclear physics, for example. You might even say we’re comparing apples and oranges. But in the context of the AI economy, where all these apples and oranges are being consolidated into AI models and made ubiquitously available to the general public for relatively cheap monthly software subscriptions, they are really all part of one big bowl of fruit salad.

Subscribe now

To make things more concrete, I’ll start with a long-winded definition. Over time, through more careful examination and reception of some tomatoes in the face, I’ll simplify it down to a more fundamental form:

Domain expertise is specialized knowledge, creative abilities, and practical skills that traditionally commanded premium prices due to their scarcity and the significant time, talent, or experience required to develop them.

This encompasses all forms of knowledge and capability that add value between raw information and actionable solutions—from technical subject matter expertise to creative expression to hands-on know-how—regardless of traditional field boundaries.

These are the very capabilities that AI systems are now consolidating and commoditizing, making them widely available at low cost and disrupting the established supply-demand economics that previously sustained expert-based business models.

Supply Chain Impact Unifies a Common Class

This broad definition is designed intentionally to include a very broad set of concepts. Similarly, the group of people who should be considered domain experts, those who possess domain expertise of some kind, will be correspondingly broad. Whether a plumber with 30 years of experience serving residential buildings on the north side of Chicago, a painter with a distinct surrealist style inspired by exposure to war zone conditions as a child, a sports medicine doctor working with professional triathletes, or a software engineer with a decade spent migrating a legacy code base into a modern architecture, each individual will have developed a unique perspective. Each will have some market for that perspective, some being larger than others. Some perhaps vanishingly small except for the right niche audience.

This industrial-scale consolidation and commoditization of expertise impacts domain and subject matter experts, IP owners, creatives, inventors, tradespeople, and really anyone with a valued point of view or niché skills born from experience in a common way. The price of expertise is being driven down rapidly by AI systems such as ChatGPT. Nearly all of these systems are trained on vast data sets of questionable provenance, which turns out to matter a lot for those whose previously monetizable work has been incorporated into those data sets without consent or compensation.1

The Supply Chain of Expertise

Consider how experts monetized their expertise in the pre-ChatGPT era2. Very complex questions required domain experts who could charge a premium for answers. The less complex the question, the greater the supply of people who could provide the answers.

With the rise of LLMs, the demand (call it “willingness to pay” if that helps) for expertise has been significantly reduced. People can just ask ChatGPT (or Claude, etc) questions for which they used to need to seek out an expert. This reduction in demand for “human hosted expertise” has an unfortunate side effect. As the demand for expertise goes down in terms of absolute dollars, the amount of supply contracts. Suppliers (experts!) exit the market as it becomes less and less sustainable for all but the most efficient operators.

If you’re skeptical about this, consider the case of Stack Overflow, an online watercooler, crowdsourced knowledge base, and one of the most commonly used online resources for software engineers in the pre-LLM era. As I learned from an article published by The Pragmatic Engineer, Stack Overflow is Almost Dead:

Note here that this isn’t a graph of the questions answered on the site, it’s a graph of the questions asked. The demand for expertise, at least in the crowdsourced Q&A format offered by Stack Overflow, has seen a catastrophic decline.

Now, to be clear, demand for the expertise being sought on Stack Overflow didn’t disappear. Buyers found in ChatGPT and other AI tools an option that served their need for expertise a tool that was significantly better. Rather than post a question and wait for a reply, one could ask a question and get a nearly instant answer. It didn’t help Stack Overflow that their entire site was scraped for training AI models, so any question already answered somewhere on the site would also be answerable by ChatGPT. If a question wasn’t answered directly, but one could assemble several pieces of existing knowledge to get an answer, searching the website would take significant time. AI chatbots could find and assemble the relevant information from within its own training corpus in just a few seconds.

Share Artificial Economics

Like It or Not, It’s Just Better (as a User)

Whether you’re a believer or skeptic on AI, it’s just a fact that the GPT-4 class LLMs all offer a vastly superior user experience for searching for information on the internet3, especially when enhanced with agentic research and reasoning functionality. Searching the internet was revolutionary as the internet emerged as an “information superhighway”. But the process of first picking the right search terms, then scouring the returned links, following them, and manually scanning websites to try to find what you’re actually looking for seems archaic in contrast to being able to ask your question in natural language to Claude and getting near immediate answers.

One reason Google is so heavily invested in (and hyping) AI is because it recognizes the threat these tools pose to their core search business. It’s why you get an AI generated summary at the top of your google search whether or not you want it. Hey, it’s better than a solid a page of sponsored links.

Supply Chain Disrupted

The incredible speed of access to information, knowledge, and expertise provided by LLM-based AI systems has had the effect of dramatically lowering the cost (time and/or money) for individuals to access it. This is not the first time the expertise supply chain has been disrupted. Just as the rise internet of the internet disrupted existing expertise-based business models with a 10x better way to acquire knowledge from the comfort of home, so too has the rise of AI with a 10x better way to sort through all that available knowledge and summarize what’s most relevant.

But beyond just “knowledge”, AI models have also been incorporating “style”. LLMs are able to reproduce the writing styles of authors with enough training material. Audio models, similarly, are showing the capacity to reproduce the styles of musicians and even the likeness of singing voices. Image and video generators are able to create stylistically dead-on impressions of artist and filmmaker styles.

Next time, I’ll get into how past disruptions in the music business compare and contrast with the disruption AI is having on the industry today in an effort to convince you, dear reader, that creatives, technical domain experts, and anyone with IP or unique perspective are in the same canoe in the AI economy and will benefit from sharing the best ways to protect the sovereignty of our expertise.

But this doesn’t mean IP rights holders and domain experts need to wait for idly for the next election. Markets are developing around this tension in the form of better tooling to prevent IP scraping (as I wrote about in my last article), but there are also brokers matching up those with desirable IP that is not easily scraped with AI labs willing to pay a premium for high quality, curated, and sometimes niche domain expertise and IP.

They Aren’t Coming to Help

If you've been hoping for regulation to curb use of copyrighted IP by AI labs, it's unlikely to happen in the US through legislation. Some court rulings have favored rightsholders, but this will create a patchwork of precedence, not any universal protection. Given the stated position of the administration along with the political and economic power held by the big AI labs, we may even see legislation to overturn these court rulings by explicitly adding AI model training to the list of "fair use" if it becomes a big enough problem for the labs.

This doesn't mean there's nothing you can do to maintain the value of your IP. There are solutions emerging in the market for protecting the value of IP and monetizing it new ways.

Assume Anything Public Will End Up in a Model

As a business owner or individual, you'll need to become the gatekeeper of your own IP. Be deliberate about what you make available online. Some assets benefit from broad reach. Ensuring web scrapers can easily consume your public sales and marketing materials and press releases may be beneficial. Consumable support documentation may help your customers serve themselves with off the shelf AI assistants rather than an in-house chat bot.

However, high-value IP that gives you a market advantage needs to be thought about differently. Paywalls are already implemented by many companies for this purpose, but in some ways they restrict access to some materials that may be desirable to expose to web scrapers through blanket policies. New capabilities like I wrote about in my previous article on Cloudflare should be considered as you decide how to keep some content available to humans but make unavailable to web scraping bots. But beyond re-establishing good hygiene for publishing on the AI-era internet, there's a bigger opportunity emerging for domain experts and their IP.

They Need Mooooooooooore.

Here's the truth. The frontier model developers (OpenAI, Anthropic, Google, et all) have already taken it all. They've already essentially used the entire internet. And…it's not enough. It's not enough for them to deliver on the generated hype. They still need more. Specifically, they need domain expert knowledge, and they need it in large amounts. This is why the market for AI model training data is exploding. See the recent acquisition of Scale AI by Meta for $13B and the surge in business for their competitors like Handshake.ai. "Handshake AI connects domain experts with leading AI labs to test, challenge, and guide how LLMs learn by providing the human judgment AI needs."

But there's a problem with this business model. I've only been in the consulting business for a year and, but over that time I've received the advice almost daily that you shouldn't charge an hourly rate. You can and should charge based on the size of the problem you're solving for the client. Domain experts are paid a decent hourly rate to provide their knowledge, but is that the true value of that knowledge to the AI labs that have billions on the line now and trillions at stake in the future?

At the discipline level (e.g. "Physics", "Law", "Medicine", etc), the general knowledge base is becoming commoditized, but new research is continuously developed and needs to be incorporated to keep models up to date. These are the areas on which large AI training data brokers are focused, and their main customers are the big AI labs.

As I wrote about before, Cloudflare sees that their customers place value on solutions that can protect their publications from being scraped for no compensation, but they’re still in the publishing business. They also value solutions that can help them sell more of their publications. Their marketplace for web scraping empowers their customers to do just that.

There are other emerging players like Protege, making markets for film and audio rights holders with data curation tools to filter, tag, and otherwise make the IP more readily consumable by labs and then brokering licensing deals. These early movers are a great signal of a broader opportunity to monetize IP before it ends up inadvertently (and without consent, for that matter) in large scale training data sets of the AI labs.

The Opportunity for Domain Experts and IP Owners

Even with all these efforts, however, there is still an enormous amount of domain knowledge that AI models can’t access today. The opportunity for those who have it is to understand first who would value LLM-powered access to this knowledge at a premium and second, how can they best capture it and package it for use in AI systems that those potential buyers would use.

While large scale, generally applicable data sets attract the interest of the AI labs. There are many small to mid-sized organizations operating in market verticals that would stand to benefit from AI tools enhanced with niche domain knowledge. The opportunity is there, but the market is still very nascent. There aren’t many people who have this kind of domain knowledge who also know how to best to curate it and package it for AI systems. Similarly, potential buyers aren’t yet aware of an immediate need beyond what they’ve already seen with off the shelf AI. But I expect that players in a few niche areas will emerge with structured and curated knowledge bases, easily integrated through MCP or other protocols as off-the-shelf agents.

Cue the contrarians who will say “the foundation models already can do all that”. They can do a lot of very impressive things, indeed, but they don’t have domain knowledge that isn’t in their training data and new domain knowledge is being developed constantly. Change is accelerating and the latest methods and techniques in many disciplines are evolving. Without a continuous process for incorporating new knowledge, eventually AI systems will fail to reflect (even a heavily biased) reality and will lose utility as they drift.

This is great news for domain experts.

If data is the new oil, domain expertise is the new gasoline.

It’s the refinement of data and experiences into useful real-world insight. A market for it will emerge as a supply chain of domain experts, curators, brokers, and integrators figure out how to best work together to maximize its market value.

We’ll get more into some of these dynamics in future articles.

But it’s also left those with their IP available on the internet with little short-term ability to prevent their IP from being consumed by web crawlers as training data for new AI models that then generate billions in revenue with no compensation flowing back to the original creators. Cloudflare’s recent actions suggest they see a unique opportunity to rebalance this dynamic between AI labs and publishers.

Cloudflare’s annoucements

Cloudflare announced on July 1 the opening of a private beta of a pay per crawl service. This service allows a domain owner to set a flat price per request for crawlers to access their content. They're also working on new tech to provide transparency between AI web crawlers and content owners, allowing web crawlers to identify themselves and give content owners control over what crawlers they will allow.

Cloudflare has also blocked all AI web crawlers from their customers' web content by default. Changing this default setting will greatly reduce the flow of free data that AI labs were using for training by simply changing their domains from opt-out to opt-in. Essentially, Cloudflare is standing up a marketplace for AI training data, putting themselves at the center of what may end up being an enormous market opportunity.

An marketplace for expertise

Training data, but more specifically, knowledge and expertise scraped from the internet, are the raw materials of large language models. If you can reduce your COGS (i.e. cost of training data) to zero, you'd be irresponsible not to do so, and so this has been the approach of the AI labs. However, the long-term consequence of this reduction in COGS is a reduction in the supply of high quality raw materials (i.e. new creations, innovations, knowledge, and expertise).

When the price of goods decreases, supply decreases. When the supply of knowledge and expertise decreases, innovation decreases. Since the release of ChatGPT, the value of many types of expertise has been in free fall, but this may be a first step in restoring some price stability around expertise in the new AI economy.

With Cloudflare’s move, a free market solution starts to emerge for how individuals and businesses can maintain dominion over their expertise. By allowing their customers options to allow, deny, or charge a per-crawl fee for this type of traffic, they’ve set up a marketplace for those seeking expertise and knowledge for model training that respects the exchange in value between the IP rights holder and the AI model vendor. Since the release of ChatGPT and popular rise of multi-modal AI models, the value of many types of expertise has been in free fall, but this may be a first step in restoring some price stability around expertise in the AI economy.

Subscribe now

Cloudflare’s unique advantage

For Cloudflare, this is very big. It represents an opportunity for them to grow from a startup darling and disruptive innovator in their market (not small already) and move upstream in the value chain essentially as an IP broker. They are uniquely position to do this because of where they sit in the tech stack today and where they aren't playing compared to their competitors (i.e. frontier model development that requires mass amounts of training data).

Solutions at higher levels of the stack, such as at the application level with paywalls and other gatekeeping methods all have undesirable effects on end users. Cloudflare handles this problem as web traffic flows through their infrastructure, a solution much more effective at blocking undesirable traffic (web crawlers) while simultaneously much less intrusive on desirable traffic (human users). There are no other players with the same combination of scale to deliver and incentive to do so.

Pressure on the hyperscalers

I expect the precedent Cloudflare's actions have set are more important than the short term impact they might deliver. They control more than 20% of the internet infrastructure market, but what about the other 80%? This is where Cloudflare has a unique advantage. Their moves will put other large infrastructure providers in a tricky position. Microsoft, Google, and Amazon all have large stakes in continuing to collect model training data with a ton of skin in the frontier model development game. These companies will be pressured by their customers for similar offerings, but have less upside to do so as it will increase costs for model training in the short term.

Rebalancing the Supply Chain of Expertise

If Cloudflare is successful in creating a market for content crawls, something many in the AI space including myself have been expecting to emerge since late 2023, I expect others will be forced to follow. For Cloudflare, this means going up market and greatly expanding their opportunity for growth. For IP rights holders, it's a first step towards being able to sleep again at night in the AI era with a plausible option for protecting one's IP while being able to monetize it in a new way.

And for the AI labs, although I expect this to be perceived initially as slowing their ability to innovate, it actually begins a critical rebalancing in the AI economy. If the value of expertise and knowledge becomes effectively zero, the slowing of innovation to follow will dwarf any short-term impacts of these market shifts. At least we can squint and see start to see pathways to a viable, sustainable, and more equitable sharing of profits of this new technology across the broader supply chain of expertise, and a functioning supply chain of expertise is fundamental for AI models to continue to improve at the pace we’ve seen over the last several years.

Leave a comment

But what’s happening in today’s economy is a bit unprecedented. What happens when expertise across nearly every domain and accumulated over millennia commoditizes at a rate faster than we’ve ever observed in human history?

The Supply & Demand Economics of Expertise

When we lower the price of expertise/creativity, we lower the supply in the marketplace at a given price. The consequence of this price reduction on creative and expert knowledge work happening across the economy due to the way emerging AI technology is being deployed and monetized is that there will be less creativity and less expertise developed.

I have to use the Mollick-ism here: I’m not talking about the future. I’m talking about things that have already happened. Take a look at what happened to Stack Overflow, one of the largest online water coolers for domain knowledge in software engineering, upon the release of ChatGPT (credit to The Pragmatic Engineer for this image):

Fewer and fewer questions are being asked on what was once one of the world’s most important marketplaces of software engineering expertise. With a reduction in demand comes a reduction in the value of the expertise on offer, which leads to a reduction in the supply of this expertise. This is a canary in the coal mine when looking at the impact of AI on value of expertise.

Maintaining Dominion Over Expertise

Businesses that rely on domain expertise and individual creators, inventors, engineers, academics, consultants, etc, all face a common problem now. So long as their domain expertise is easily consumed by web scraping bots or freely given away through terms of use on social platforms, they will struggle to maintain price premiums.

Of course, domain experts have experience they pair with knowledge to bring differentiated value to their company, clients, or craft. The sum of the value they create is typically not all documented in machine-readable form. But now is the time for all businesses and individuals to get more intentional about how their expertise will co-exist in a world with ubiquitous AI systems that are designed to extract this expertise and monetize it for someone else.

My intention for this newsletter is to explore the supply chain of expertise as it is disrupted by AI and to provide insight and practical advice for individual experts, creatives and businesses that want to maintain dominion over their expertise, domain knowledge, creative style, and experience.