When discussing Crypto, the largest narrative after Memecoins and Stablecoins lately has been decentralized AI (aka DeAI), and in the last couple of weeks, agents in particular. As with any new relevant narrative, there is a debate about how much substance is behind the hype.

The negative side sees only a negligible overlap between crypto and AI, criticizing that AI (or Agents) times Crypto is just one of those ideas that wants to generate hype by simply mashing two hyped topics together without a relevant reason. History will consider this as one of the worst takes ever.

Sergey Gorbunov

@sergey_nog

On the intersection of AI & Web3.

Thread

450

13:23 • 22 Nov 2024

But even if you believe in a larger overlap, I consider this to be the wrong mental framework. There have been many intersections of technologies that went to zero, and an even larger set of intersections that created small niches that were just a fraction of their parent technologies. But there is a third scenario that unlocks something more groundbreaking: Synergistic technologies.

Synergistic technologies are two technologies that are so complementary in their nature that they massively expand each other's TAM. In very left-curvian terms: The sum is larger than its parts.

One example of two synergistic technologies is modern smartphones (aka iPhones or iPhone derivatives) and social media. Both drove adoption of each counterpart: While Facebook and other social networks existed pre-iPhone, they significantly expanded all usage metrics once they had a suitable mobile platform. On the other hand, Apple can achieve such high ARPUs because people want great hardware that allows them to share high-quality content with their friends.

Once these synergies played out these two technologies create strong gravitational pulls towards each other: It is nowadays impossible to imagine a social network that does not run on mobile phones and it is equally impossible to think of a mobile computing platform that wont interact with social networks.

I believe the same will be true for crypto and AI agents. I argue that this multiplicative effect is so strong, that the majority of economic value that agents capture, create and distribute will be on crypto infrastructure. At the same time, I argue that the majority of transactions on crypto will be agent-driven.

In order to create a strong gravitational pull between two technologies, each technology must have beneficial characteristics. It is not sufficient for one technology to unilaterally benefit the other; rather, the synergy effects must be reciprocal in order to ensure a massive TAM expansion. For crypto and AI agents, there are some clear contenders for this:

In its core, AI agents are systems designed to perform tasks or achieve goals autonomously. They operate by processing inputs from their environment and making decisions based on predefined algorithms or learned experiences to maximize the fulfillment of their objectives. Agents are goal-oriented and agnostic about the methods used to achieve these goals, focusing solely on the end results.

This agnosticism is the essential multiplicative effect for crypto. In marketing, there is this decade-old paradigm of how products get adopted: Innovators, early adopters, etc. This adoption curve is essentially driven by awareness and bias against the product. It takes time to ripple through the channels people use to get information (how long does it take until people hear about XYZ) and often even longer until they like it (bias).

Agents completely squash this adoption curve because they are not restrained by human attention and social or emotional bias. It takes years to convince many people over 30 to use crypto, but an agent will just unemotionally use it based on the input and the model data.

When people evaluate crypto, they usually tend to overfixate on the loud noise, i.e., speculation, memecoins, and co. Let us entertain the position of being extremely cynical about crypto: Even if you believe it does not provide any real additional value when compared to the general open finance space (aka FinTech infrastructure), it remains the largest open-source fintech infrastructure stack that allows for the widest variety of use cases.

This is hugely synergistic for agents, because they are in need of interfaces that allow them to facilitate transactions in order to achieve their goals. Not only has the Crypto VC industry poured billions of USD into the development of open-source infrastructure, but they have also found a way to monetize this infrastructure pre PMF via token launches. This allowed the crypto industry to keep building on infra even in the deepest bear markets. As a result of this, crypto tech infra vastly outsizes and outcompetes web2 open finance infra.

Crypto is pseudonomynous by nature. There has always been a conceptual disconnect between wallets and users when compared to the analogue in traditional finance (accounts and users). Many people have anonymous secondary or tertiary throwaway wallets in a way they do not have in the traditional banking system. That is because each banking account is onboarded via KYC.

L1s and L2s often measure their traction in monthly active wallets instead of real individual users (Mostly because they simply cannot tell the difference). In crypto, it simply does not matter who you are interacting with as the vast majority of popular applications remain completely permissionless.

This creates synergy both for crypto and agents. It is much harder for agents to navigate the regulatory environment of traditional financial infrastructure (and in most cases there is no real regulatory precedent). Meanwhile crypto from the get go speaks the language agents natively understand: Two parties interact by a predefined logic (code is law) and transactions are verified via cryptography. Because regulation will take some time to iron out, crypto will be the perfect experimentation space for agents in the meantime.

There are two axes along which crypto and agents will expand their TAMs reciprocally. The two different ramp ups are benefitial for the crypto landscape, because it somewhat hedges the speed of the general adoption of crypto native Defi products with infiltration of stablecoins into traditional web2 payment rails. It also derisks for the case of larger agent models from the web2 hyperscaler world dominating the agent narrative. The two axes are:

Native agents are born on-chain and built on crypto infrastructure. There has been plenty written about popular launchtools such as eliza, Coinbase's Agent SDK and others, and more chains will launch their own frameworks (see Solana). There will be a similar war for Agents that we had for devs and users, so expect incentivisation programmes (aka grants, accelerators, airdrops) to skyrocket for native crypto agents.

Agents live on a spectrum in terms of their autonomy. Some will keep running autonomously from an initial human-set goal (e.g. Terminal of Truths) and others will translate human intent to transactions. The first category will expand TAM by essentially onboarding more wallets on-chain, the latter will make crypto less complicated, more powerful for users and help onboard more human users on crypto applications. There is already a number of LLM first interfaces to facilitate on-chain transactions (see Bitte or NEAR's AI assistant) and I do believe that this will become the primary interface how crypto transactions will be triggered within the next few years.

There will be many agents that are not natively born on-chain, developed by web2 companies that will still end up moving their funds via crypto rails. The primary reason for this is major web2 FinTech companies starting to adopt stablecoin infrastructure for payment rails. And so I believe many web2 agents will move crypto without necessarily intending to or knowing about it. An example of this is Stripe: The company recently announced their Agent's toolkit, and also intensified their crypto efforts with product development and acquisitions. It is reasonable to expect that they will try to push as many of their transactions.

While this usecase may not seem as interesting from an experimentation perspective, the volume it creates could still massively expand crypto expand DeFi TAM. Agents will find more use in microtransactions (because they are not limited by attention, both lower and upper bound), and traditional FinTech companies will push agent transactions onto crypto rails to keep costs down.

Whenever a new technology (or a combination of two) hits the market, different verticals within that technology have different adoption timelines. The same will be true with agents using crypto.

The agent narrative initially kicked off with agent-driven memecoins, driven by the likes of Truth Terminal or Zerebro. This led to the creation of frameworks and Agent launchpads, with many agents being launched today still being in the meme or social category. An example for this is Clanker, which is a Farcaster focussed agent that allows you to launch memecoins. More of these platforms are in development, and as we are in the bull market, I expect the noise around these to remain high.

Additionally, we will see more long-lasting vertical usecases picking up steam in the next couple of months, particularly in the DeFi sector. For instance, Almanak allows you to use and develop Defi-specific agents.

Indirect Agent x Crypto TAM expansion will creep up steadily, as web2 and institutional adoption of crypto payment rails (in in the form of stablecoin settlement) continues to grow.