< Back to Writing
June 9, 2023
AI, Crypto, & the Return of the Bazaar

AI, Crypto, & the Return of the Bazaar

Part I: Cryptonetworks & the Bazaar 

In 1999, software engineer Eric S. Raymond published his seminal essay on open source software and the hacker ethos, The Cathedral and the Bazaar. The essay outlines two contrasting models of software development: the Cathedral model, characterized by a top-down, meticulously crafted approach and the Bazaar model, characterized by decentralized, open, community-driven building. Ultimately, he argues, the Bazaar model creates better code by having tighter cycles of iterative development, more eyeballs to catch bugs, and more ongoing stakeholders to future-proof that code against emergent complexity. 

Cryptonetworks are distributed open software protocols so, not surprisingly, OG crypto enthusiasts have often cited Raymond’s framework over the years to explain why crypto’s eventual victory is “pre-programmed” (read: inevitable). 

And yet, fast forward to a bear market in the era of TikTok, Facebook, Twitter, Google, Microsoft, Apple (towering Cathedrals by anyone’s standards), and Raymond’s 1999 essay reads more like an unrealized momento of early hackerdom than a prophetic manifesto of an open-source future. Notable exceptions to the dominant Cathedrals persist – email’s SMTP protocol, the Internet’s TCP/IP and HTTP and DNS protocols, ongoing use of the Linux kernel, among others – but until the launch of Ethereum’s ERC-20 fungible token standard in 2017, open networks were hamstrung by lack of coordination incentives. Open networks could simply not compete with the funding and blitz-scaling strategies of centralized networks.

In a 2016 essay, The Golden Age of Open Protocols, USV’s Fred Wilson explains, “One of the problems we have had in tech is that there aren’t large monetary incentives to create and sustain open protocols...If they are open,” he writes, “they cannot be easily monetized by traditional means”. 

A16z’s Chris Dixon fleshes out the dynamic further in his 2018 blog post, Why Decentralization Matters, arguing that centralized platforms have historically outperformed open protocols by doing the following: (1) Raise venture dollars on the promise of outsized future value capture, (2) Use those venture dollars to subsidize a great product for users, (3) As the platform scales, extract value from users (and builders) to return outsized multiples on invested capital to investors. Users lack an at-parity alternative, so the model works and attracts more capital to rinse and repeat at the long term expense of platform participants. 

Graphically, Dixon represents the resulting platform dynamics as follows:

07lrwGIDbAYk6q7zG

Wilson, Dixon, and other fund managers saw early on that cryptonetworks could offer a different business model, one where platforms can be built as self-sustaining protocols and where, as a result, the user can be treated as both builder and owner. They weren’t wrong, as evidenced by consumer, institutional, and developer adoption we’ve so often covered in past letters. 

What’s Been Missing? 

But bear markets and non-zero interest rate environments remind us that better business models and battle-tested code aren’t enough; on-chain behaviors and off-chain conversations remind us that for resilient at-scale adoption to accelerate – for the visions of Raymond, Wilson, and Dixon to be realized – one or both of the following must also be true about the open network offering:

  1. If disrupting a Cathedral’s solution, the open network offering must be at-parity or better than that of the legacy player. This seems simple, but builders and investors alike often behave like they can change the hierarchy of what the user wants or needs. If onboarding for your consumer social app is confusing or the experience is bad, users simply will not see how great it is to have an immutable or composable social graph. If your game is not fun to play, ownable assets or verifiable game rules will not matter when venture-funded incentives dry up. 

  1. If solving for an emergent problem or behavior (ie: the user is organically experiencing a problem with no solution yet offered by the Cathedrals), that problem must be one where the properties of crypto-enabled solution are uniquely suited for that new job

With that, we look to the current environment and ask, “Are we accelerating towards at-parity experiences? Are there new problems that are best solved by crypto tools? And most importantly, how large and lasting do we expect those opportunities to be?”

Part II: Resilient Demand & Reigniting the Bazaar 

Armed with the framework above, we’re paying special attention to 3 themes that we believe will coalesce as unintentional tailwinds for a next wave of resilient crypto adoption:

  • Generative-AI is democratizing the creation of at-parity software. Software is becoming commoditized, and, as Cathedrals lose their software moat, open platforms that leverage cryptonetworks will become viable at-scale alternatives.

  • Generative-AI is exposing new problems that crypto is uniquely suited to solve. Consensus reality and trust on the Internet are breaking down. Cryptographic truth will be part of the solution. 

  • In a globally-networked world, legacy “network effects” are more vulnerable than ever. New networks can be memed into or out of existence in ways they never could before (hint: SVB, OpenAI, etc…). 

I.  Generative-AI is democratizing the creation of at-parity software.  

ChatGPT crossed 100 million users and over 13 million unique users per day within 2 months of launch. One month later, Microsoft reported that a whopping 40% of code submitted by developers using GitHub Copilot was AI-generated and unmodified. For Java, that number was even higher, with AI generating over 60% of code in editors where it’s used. And while AI co-pilots are exciting, we’re already seeing autonomous AI agents like AutoGPT, HuggingGPT, or BabyAGI and others empowering even non-technical developers to create basic software.   

The Cathedrals of web 2.0 may have dominated because they could raise capital, subsidize a best-in-class product, capture market share, and then extract value over time from users who lacked at-parity alternatives. But that’s changing: leaner, less capitalized, more distributed builders are creating software and digital content that, even a year ago, would have seemed possible only for their Cathedral counterparts. When anyone can create software, the software industry starts looking more like a software ecosystem and more like Raymond’s Bazaar – only this time around it’s equipped with coordination mechanisms in the form of crypto rails. 

A fun example is imagining a sufficiently decentralized and more community owned Uber; historically, this was a laughable example, an iconic example of the 2017 ICO bubble – a small distributed team being able to compete with thousands of Silicon Valley engineers! In 2023, it’s less far-fetched to consider an open source matching algorithm could soon get to parity with Uber’s app. When (not if) this happens, will drivers stay with the more extractive platform? Will riders stay with the more expensive platform? Ask your uber driver.

Of course at the same time, teams are also leveraging AI to more efficiently write and audit smart contracts, pressure test code and architecture vulnerabilities, make blockchain data more human-readable/conversationally searchable, simulate transactions to protect against nefarious contracts, and more. Crypto native rails are getting friendlier to use too. 

II.  Generative-AI is creating new problems that blockchains are uniquely suited to solve.

AI is creating new push-functions towards crypto, new problems that crypto is uniquely capable of solving where there is no clear at-parity alternative. 

To name a few:

  • Need for Cryptographic Truth - cryptonetworks are built and strengthened by the adversarial nature of the Bazaar.  To be “trustless” means that a user doesn’t actually need to trust their counterpart because there’s mutual trust in the blockchain or code. This concept when applied to internet money or speculative jpegs might seem like a nerd trap or toy, but this year we’ve seen AI create large scale questioning of what consensus reality really means. Authentication of digital content – be it a photo or a tweet or a news story – is getting increasingly adversarial, and we’re confident cryptographic authentication and public blockchains will be a major part of the solution. 

Left: AI-generated Pope in Puffer Jacket meme that took the world by storm. Right: AI-generated Pentagon explosion photo that caused a same-day 30 point swing in the S&P. Still think this is a toy, anon? 

  • Proof of Humanity - narrowing in on one particular aspect of truth, an underappreciated quality of the internet to date has been that, generally we believe a human to be behind a given content or action. How will we know that the tweet we’re reading, the video we’re seeing, or the voice we’re hearing in real-time is authored by the human we believe it to be or a human at all? Increasingly there’s even need for proof of humanity in the training of AI models, with a Canadian research team publishing findings in May suggesting that “use of model-generated content in training causes irreversible defects in the resulting models”. The question then becomes, is a centralized system – presumably state-controlled or international organization – the best path forward for your global online identity?  Or do we opt for a decentralized, privacy-preserving, and immutable framework? 
  • Autonomous Agent Economy - as autonomous AI agents become more viable, we’re seeing creators experimenting with crypto wallets for their bots or even controlling ownership of those bots using NFTs. Altered State Machine, for example, is building the tools for creators to build AI bots that can operate autonomously in-game and become cash-flowing assets tradeable as NFTs.
  • Digital Asset Authenticity - zooming out, when anyone is empowered to quickly create digital assets, whether scarce or abundant or valuable or free, we believe open network infrastructure for tracking those digitally-native assets will become only more relevant.  

III. Network moats are cracking

Today the term “Network Effect” implies an almost god-like indomitability, but when originally coined in 1908 by AT&T chairman Theodore Vail, the term simply highlighted that the value of a network to a given user grows as the size of the network grows. In 1980, engineer and later inventor of Ethernet, Robert Metcalfe, added that the financial value of a network would be proportional to the square of the size of the network (“Metcalfe’s Law”). The math is fuzzy and has been subject to plenty of debate, but it seems safe to say that, at least directionally, the value of a network to the user and to the market – be it early telephone infrastructure or cryptonetworks like Ethereum – is indeed related to its size. 

What’s less clear in 2023 is whether the implied defensibility of large networks is as strong today as it was one hundred years ago or even five years ago. Networks that have physical infrastructure (e.g. Verizon, AT&T, T-Mobile) or proprietary inventory (e.g. Airbnb, Social Networks) have been among the most durable to date.

But what happens when switching costs go down, when information exchange hastens, and when at-parity alternatives exist? This point perhaps warrants its own deep dive in a future letter, but we believe that in an increasingly networked world (i.e. consumers are networked and exchanging information about platforms outside of the platforms themselves), incumbent platforms are quickly becoming more vulnerable to displacement. 

Looking at examples from just this year – the 48 hour flight of depositors from Silicon Valley Bank, the unprecedented pace of OpenAI’s scaling, the subsequent rise of open source AI alternatives like Stability AI, the news that Apple will be forced to allow sideloading of non-App Store apps on all new EU devices by 2024 – when at-parity products emerge, when switching is as easy as thumbing a mobile wire or setting up a new account, and when platform shifts can accelerate across Twitter, email, Telegram group chats, substacks and any number of other channels…we enter a world where Network Effects can crack in surprising ways. 

We believe cryptonetworks will underpin many of the largest open software protocols of this next era and that the Fund is in a tremendous position to ride that next wave of resilient at-scale crypto adoption.