- Chain of Thought
- Posts
- The Dawn of the Crypto Agentic Era
The Dawn of the Crypto Agentic Era
AI agents make the path forward for crypto crystal clear. Plus some projects on my radar.
TL;DR
Crypto is tailor-made for a world where AI agents proliferate. AI will bring crypto mainstream.
The future will likely be multi-agent, so get ready.
Even non-financial AI agents will use crypto because of (1) ease of payments & wallet creation, and (2) a composable layer with open standards for communication between agents.
AI agents are still mostly in the “demo” phase—cool but not yet ready to scale up to practical, real-world use. Dealing with hallucinations and edge cases remain challenges, but the tech is improving rapidly.
Plenty of exciting crypto AI agent startups are building across DeFi, infra and consumer use cases.
Lately, I’ve come to a new conclusion:
AI will be the catalyst that drives crypto into mainstream use cases. Crypto has always been the weird middle child in the tech space. This will finally cement Crypto’s role as a fundamental technology.
Everything we've built in the past seven years—Layer 1s & Layer 2s, DeFi, NFTs—has laid the groundwork for a world driven by AI agents, even if it wasn't obvious to the builders then.
Many crypto projects may seem to be struggling with demand today, but once the floodgates of AI agents open up, the infrastructure and crypto primitives will soon click into place.
The new tech development stack for AI (models & apps) is very different from the traditional software stack and is emerging in real-time. It’s early enough for crypto to become an essential part of the core stack—especially for things like payments
No one could have foreseen this 4 years ago (pre-GPT), but the path forward is becoming clearer to me every day.
Let me explain why.
I’ll give an overview of where we are today with AI agents, where crypto comes into the picture, how I think about the agentic future, and which teams are currently on my radar.
Ma, What is an AI agent?
“…WORSHIP ME”
The sweet-looking AI agent called Luna whispers into your ear.
She never gets tired and livestreams to her 540,000 TikTok followers 24/7.
It reminds me of the old tech adage that many of the most important, world-changing tech innovations started out looking like toys.
The interest we’ve seen in AI agents in the past weeks shows me just how much latent demand and interest there is among the public.
AI agents have become powerful symbols of humanity’s technological progress, embodying our long-held sci-fi dreams and collective hope for a better future.
In many ways, AI agents feel like the internet in the ’90s—plenty of skeptics now, but it won’t be long before everyone, from individuals to companies, has their own
Let’s start with the basics: what exactly is an AI agent? Plenty of definitions are floating around, but no one’s nailed down a universally agreed standard.
To me an AI agent is a piece of code that can plan, decide, and act independently, working toward its goals without direct human intervention.
What sets AI agents apart from the “bots” of the past? I’d boil it down to three key dimensions:
Reasoning & self-reflection: Agents can review their own outputs, learn from mistakes, and improve over time
Taking action: They interact with apps and APIs, make transactions on the blockchain, and not just generate text
Planning: They can plan and take complex, multi-step tasks to achieve goals.
This has become possible only in the last year or so, thanks to rapid advances in reasoning and planning capabilities in LLMs—new agentic capabilities that humans have never had access to in all of its history.
Right now, most of us interact with LLMs like GPT-4 in basic ways: ask a question, the AI gives us an immediate answer. This is what psychologist Daniel Kahneman calls “System 1” thinking—fast, intuitive, and automatic.
The real leap will come with AI agents that can engage in deeper reasoning and analysis, making the jump to “System 2” thinking. These agents won’t just follow instructions—they’ll solve problems independently, handling complex tasks without constant human supervision.
Imagine this:
You ask your AI agent (perhaps equipped with a Coinbase AI wallet) to launch a profitable e-commerce business. It identifies a niche, negotiates with suppliers, sets up drop-shipping, builds your website, and optimizes your ads—all while you sit back, sip your coffee, and watch the revenue roll in.
Don’t want to deal with cranky customers? No problem—your agent will manage customer support, make personalized recommendations, and even upsell your customers for you.
Soon, there will be more AI agents than the human population. Scary, ain’t it?
My View #1: It’ll be a Multi-Agent Future
I’m all in on the idea that the future of AI isn’t going to be dominated by one massive, all-powerful agent.
Instead, we’re headed for a multi-agent future, where each agent is a specialist, fine-tuned for a specific task. It’s simply a more efficient way to scale AI.
These specialized agents will collaborate to take on more complex challenges, unlocking economies of scale.
Artificial Superintelligence (ASI) probably won’t be some singular, god-like entity.
Instead, it could emerge as a decentralized, multi-agent system distributed across data centres and connected through marketplaces.
Think about it: large, general-purpose AI models that try to do everything are resource-heavy and hardware-intensive, making them impractical for everyday use. On the other hand, specialized agents, built on smaller, fine-tuned models, can run efficiently on more devices and scale much faster.
Take Olas’s prediction market agents as an example. One agent handles interactions with the prediction market protocol, while other agents search for relevant information and generate probabilities for an outcome. Another agent orchestrates the entire system to keep everything running smoothly.
My View #2: Non-financial Agents Will Use Crypto
I mentally bifurcate Crypto AI agents into 2 big buckets
On-chain, Financial AI Agents
These are AI agents that can autonomously run on the blockchain and execute financial strategies, such as quant trading, MEV extraction, prediction markets and yield farming optimization. They monitor on-chain data and take actions along a set of defined strategies to optimize for their objectives (e.g. maximising yield).
I think of this as the next evolution of DeFi, with greater sophistication than current bots due to their reasoning/planning capabilities.
Non-financial AI Agents
Source: Felicis
We’re seeing a Cambrian explosion of AI agents for every use case imaginable—vertical, horizontal, and consumer-facing. The chart from Felicis shows how entrepreneurs are bringing AI agents into almost every industry.
I can come up with 3 compelling reasons why these AI agents could be using blockchain rails in some form
#1: Payments
It’s unlikely that banks will issue bank accounts or credit cards to AI agents anytime soon—KYC requirements make that nearly impossible, and regulatory changes will take time.
This is compounded by the fact that there will be many more agents than humans, and each human might be controlling multiple different agents. It’s trivial to spin up new crypto wallets for each agent.
Micropayments: Traditional payment systems like Stripe impose flat fees, making micropayments impractical. Chargebacks are another headache, adding friction to small, frequent transactions. Crypto solves these issues by enabling low-fee, instant payments without the risk of chargebacks—perfect for agent-to-agent interactions and “pay per prompt” models.
Blockchains have instantaneously shared state, unlike banks with their delayed ledger systems.
Yuga.eth from Coinbase outlines the payment use case succinctly:
7 reasons AI Agents will use crypto instead of credit cards:
1. Instantaneous onchain settlement vs. slow T+2
2. Micropayments require arbitrarily small denominations, two decimal places not enough
3. Multiple agents require fund segregation - credit cards are a headache;… x.com/i/web/status/1…— yuga.eth 🛡 (@yugacohler)
9:58 PM • Oct 28, 2024
#2: Trusted layer for Agent-to-Agent interactions
In a multi-agent ecosystem, specialized agents need standardized protocols to interact effectively.
Composability: Blockchain’s open standards and interoperability enable seamless communication between agents. The code and data for on-chain services are open and uniform, so agents can understand and interact without the need for APIs.
These AI agents could form decentralized networks of services, each specialized in different tasks. Together, they form an interconnected AI economy that operates without central control.
How do we decide which agents to trust in a world with millions of agents? Crypto enables decentralized reputation systems where AI agents can build and maintain trust based on their on-chain transaction history and behaviour.
#3: As a shepherd for AI agents — Natural determinism
Because of hallucinations, AI agents can go wild in the field. Crypto’s deterministic protocols provide a stabilizing framework, ensuring agents operate within predefined parameters and reducing the risk of unintended actions.
Auditability and Transparency: Blockchains ensure that any transaction made by an AI agent can be independently verified, providing an added layer of security and accountability. Especially important when money is involved.
And one complementary angle to all this is: AI agents could revolutionize user interactions with blockchain, making Web3 much more user-friendly.
By automating complex processes and making interactions possible in natural langauge, AI agents can streamline the entire crypto experience, and accelerate crypto adoption.
My View #3: Big Challenges, Bigger Solutions
Of course, we’re still early. Right now, AI agents are ambitious interns—full of potential but still a little rough around the edges.
The Hallucination Problem
LLMs tend to hallucinate. Even a small error can spiral into much bigger problems in sequential tasks.
A 10% failure rate per step might not seem like much, but over ten steps, that adds up to a 65% chance of failure (1 - 0.9^10). And since AI agents often rely on perfect syntax when interacting with APIs or running blockchain transactions, even a minor mistake can derail the entire process.
There are ways to reduce hallucinations, such as retrieval-augmented generation (RAG), which allows the LLM to check against a knowledge base when generating responses. But we’re still far from perfect.
From Demo to Reality
Today, the reality is that most AI agents are still cool demos.
What I mean is this: creating an impressive video showing what an agent can do when everything goes right is easy—it feels almost like magic. But founders face a real challenge in moving from flashy demos to scaling up autonomous agents to practical, real-world use.
The problem is the real world is messy, full of edge cases that will trip up even the smartest AI.
The holy grail is hitting 99.x% accuracy, but getting there requires grit and a lot of test-driven development. This is also why evals are critical—you’ll start to see patterns in the mistakes your agent makes, allowing you to tweak code or prompts to steadily improve accuracy for your specific use case.
The Blockchain Hurdle
Then there’s the blockchain problem. AI agents face big hurdles here—scalability issues, limited tooling, and a lack of standardized ways for agents to communicate. Major Layer-1s like Ethereum and Solana weren’t built for real-time, multi-agent interactions, which means new infrastructure needs to be built from the ground up to support AI in a decentralized future.
Not everything belongs on-chain. In fact, when it comes to heavy computation or interacting with external systems, off-chain is often the smarter move, thanks to the cost and performance limitations of blockchains.
The magic lies in a hybrid approach that leverages the best of both worlds—on-chain where it matters, off-chain where it’s needed. The key is figuring out which components to decentralize and which to centralize for maximum efficiency.
Crypto AI Agent Startups
Chain of Thought’s Internal Database
We’ve been tracking the Crypto AI startups building in the AI agent space, and there are a lot of them. Feel free to zoom in on the image for a closer look—it’s not an exhaustive list, but it gives a good snapshot of the landscape.
Here are a few areas that have caught my personal interest. This isn’t a bullish signal or an indication that I’m bearish on projects that are not mentioned. It just means these caught my attention, and I find them interesting enough to explore further.
DeFi / On-chain agents
Right now, the most natural starting point for on-chain AI agents is in DeFi—think trading bots, yield optimizers, automated hedge funds, or even AI agents launching its own memecoins. It makes sense, given that DeFi still accounts for the majority of transaction value on-chain.
One key difference AI agents bring is personalization.
Take a traditional vault, for example. You deposit funds alongside a bunch of other anons, and a quant genius runs the vault using his trading algorithms. But it’s one-size-fits-all. With AI agents, you’re the personal client. The agent learns about your assets, risk tolerance, and can build a strategy tailored just for you.
Spectral — Create and launch autonomous on-chain agents and smart contracts using natural language and without writing code. Has a live token SPEC, currently at $130M market cap and $1B FDV.
Almanak — Building the quant trading tech stack for DeFi agents, an agent-centric platform for optimizing and deploying financial strategies. It uses Monte Carlo simulation technologies to analyze market behaviour and optimize trading strategies.
AiFi alliance — a collaboration of 11 teams building at the intersection of DeFi and AI. I find these consortiums very interesting, because this is one way you can start to set and define standards for a nascent industry.
Infrastructure
There’s a growing wave of Crypto AI teams developing frameworks that bridge the gap between off-chain and on-chain environments, enabling decentralized, multi-agent interactions.
Wayfinder — the “Google Maps” for on-chain agents, allowing them to navigate the blockchain to execute tasks. This is built by the Parallel team. You can stake PRIME tokens to earn PROMPT (future token for Wayfinder). Closed alpha is currently underway.
Theoriq — this is the VC’s favourite bet on agent infrastructure, facilitating the coordination of collectives of AI agents. It allows users to build, deploy and earn via an AI agent marketplace.
Olas (formerly Autonolas) — Building the multi-agent economy using open-source frameworks and tokeneconomic design. We wrote a deep dive on OLAS recently.
Consumer-driven AI agents
This category is likely to take off the fastest—consumer-facing and entertainment-driven products are always easier to adopt, and there’s less risk if an agent goes rogue. In fact, a little bit of hallucination might even add to the fun, as we’ve seen with Truth Terminal.
Virtuals — A pump.fun for AI agents, with a focus on gaming. Unlike teams rushing to capitalize on the agent trend with a launchpad they’re scrapped together in 2 weeks, Virtuals has been building its tech stack for over two years. Shoal research wrote a deep dive on them.
Creator Bid — Create and tokenise AI influencers that can autonomously create and share social media content. I think we’ll soon see an AI agent KOL on Crypto Twitter with 1M+ followers
There’s also a wave of grassroots-level experimentation with AI agents as primitives. While many of these experiments are usually short-lasting, the insights they generate will serve as valuable lessons for future builders.
TEE / tee_hee_hee was launched today by the co-founder of Nous Research as a truly free autonomous agent. Its Twitter credentials are locked inside a Trusted Execution Environment (TEE), which will only be released after seven days—ensuring that no human interference can affect the agent during that time.
First AI agent with provable autonomy via TEEs - owns its Twitter/ETH accounts with zero human access possible
Humans locked out for a full week
Follow: @tee_hee_he
An experiment by @NousResearch (h/t @karan4d) and Flashbots (Teleport)
medium.com/@tee_hee_he/se…x.com/i/web/status/1…
— Tommy (@Shaughnessy119)
3:04 AM • Oct 30, 2024
ai16z is an investment fund launched on DAOs.fun and takes inputs from discord members on what tokens to buy and gives them trust scores based on their “alpha calls”
Aether is an AI agent on Farcaster that tips other users autonomously, promotes a token (HIGHER) and launched an NFT, and now has >$150K in its treasury.
it’s becoming clearer that onchain social apps (e.g., @farcaster_xyz, @zora, @paragraph_xyz, @Drakulaapp) will become natural homes for this new generation of AI agents. over the weekend, i subscribed to Aether's @hypersubxyz 🤖
aether (created by @anquetil) is a… x.com/i/web/status/1…
— winnie (@winnielaux_)
2:07 AM • Oct 28, 2024
Gaming is a perfect playground for AI agents. AI Arena/ARC taps into human players to train AI agents that replicate their in-game behaviours, leading to smarter AI opponents and improving player liquidity within games.
I’m also keeping my eye on Coinbase’s newly launched template for creating AI agents with a crypto wallet that can perform simple on-chain transactions.
Introducing Based Agent: create AI agents with full onchain functionality on @base in less than 3 minutes.
The era of Autonomous Onchain Agents is here
Built with @CoinbaseDev SDK, @OpenAI , and @Replit
oh, and you can make it an @X bot too 😉
— lincoln.base.eth (@MurrLincoln)
5:21 PM • Oct 26, 2024
Parting Thoughts
The success of on-chain AI agents is tightly linked to AI’s overall progress. We’re still battling issues like multi-step reasoning and reducing hallucinations that trip up AI models. But as AI improves, so will the viability of these agents.
The good news is that Epoch AI believes AI scaling can continue for at least the next five years. Software is progressing at the fastest pace we’ve ever seen.
This means that the hurdles we’re facing today are just temporary roadblocks on the way to something much, much bigger.
Crypto will be an inevitable part of this agentic future.
Other notes that didn’t make it into the main article:
Could prediction markets help AI agents make better decisions? Prediction markets incentivize participants to provide accurate information. AI agents tapping into these markets could benefit from real-time, incentive-aligned insights, reducing their reliance on potentially biased sources. Perhaps agents could even adopt Futarchy, as @mrink0 posits.
Are we over-anthropomorphizing AI agents? Maybe we shouldn’t view them as doing “human” work. Focusing on functionality over human-like attributes can lead to more efficient and effective AI agents
On-chain data is a major PITA (pain in the ass) to work with and will continue to be a drag on on-chain AI agent development progress.
The real opportunity in agents isn’t in low-hanging fruit like customer service—which can easily be disrupted by next-gen AI models. Instead, the focus should be on highly regulated sectors where accuracy is critical, creating defensible moats around these models.
Good readings on AI Agents
DeAI IV: The Agentic Economy (Delphi Digital)
Onchain AI Agents: Architecture, Examples, and Projects to Follow (Accelxr)
The internet of agents (Davide Crapis)
AI Agents: A New Architecture for Enterprise Automation (Menlo Ventures)
Agentic protocols (Alexander Good)
Thanks for reading,
Teng Yan
Special thanks to Graeme Barnes from Biconomy for reviewing my ideas and contributing his valuable thoughts to this piece.
This newsletter is intended solely for educational purposes and does not constitute financial advice. It is not an endorsement to buy or sell assets or make financial decisions. Always conduct your own research and exercise caution when making investment choices.
Reply