Build something agents want

Build something agents want

The crypto world is buzzing with a new trend, AI crypto agents.

Agents have codified non-deterministic personalities. They access the X.com timeline, news, discord, and forecaster. They use these socials as users (read, post, reply) and interact with defi protocols.

They are already sponsoring each other's posts with tokens and buying services on their own. The value-chain of agents is forming, along with the first cases with true product market fit.

Currently, the market cap of agents - tracked by Cookie3 dao - has reached 5.5b$. Foundations, VCs and protocols are all working on an agent strategy.

We are still early, but agents are here to stay.

What are AI agents in crypto

AI agents consume information, interact with users via clients (social media), and manage crypto assets. Each agent is created by a prompt, a sort of mandate that usually define their personality, tone of voices, rules, objectives and constraints.

The agents value chain

A first stab at agents value chain

Let's double click on frameworks and launchpads, the most interesting pieces of the value chains:

  • Frameworks: open source code to launch agents in minutes. The Eliza ai16z repo - not affiliated with a16z crypto - is gaining popularity among developers. These frameworks aim to create an agent standard. It is not clear how they will get value in the long term (the expectation is that ai16z will receive tokens from ai agents). Thousands of agents have been already launched with Eliza.
  • Launchpads: Agents launchpads are the fastest way to tokenize an agent. In the case of Virtuals (already at a market cap greater than 1b), users can launch agents on a bonding curve. As the market cap increases over a certain threshold, agents "graduate" and access social media and soon other clients.


ai16z is well positioned to become the React of ai agents


Agent types

Mapping out agent types is hard, as agents are fluid. They can transform themselves and jump to other categories.

Meme/Community Agents

These are viral agents with strong community engagement and personality. This is the first kind of agents - GOAT is arguably the best example.

Companion and IP Agents

These agents can be your assistant, friend or girlfriend/boyfriend. These agents can develop and tokenize specific IPs and create NFTs and art. They market to specific niches, like anime. Many NFT collections will see these agents as an opportunity to revamp and reengage their communities.

An example is Misato, developing an ecosystem of agents for lovers of manga with explicit content.



Utility agents

These agents provide some sort of value to their holders or followers. We are currently seeing a lot of agents specializing in market insights on specific crypto verticals (defi, agents, nfts, polymarket). These are the agent equivalent of KOLs, and their mindshare on X is growing.

These are content creators that watch the market and shitposts 24/7. This is the most popular category by number of agents and market cap as of now (Dec 2024). The most popular agent in this category is aiXBT.


Example of aixbt shilling a coin

Other agents provide utility for specific verticals, like gaming. The virtual team is building a framework for AI powered game characters.

Public good agents

This is my favorite category. These are agents that pursue public good missions.

Let's explain this category by looking at an example, Freysa. Freysa is a real-time experiment on AI security. An LLM is given some money to protect, and users have to get them by interacting with Freysa. Every question has a cost that grows exponentially, therefore increasing the total payout.

So far, Freysa has done two iterations of the challenge (the first one arrived at a payout of 50k, the second one at 12k) - here's the last message that won:

Playing with CLI instruction was the winning approach in both of the Freysa challenges


Current limitations

  • Data: Agents are only as good as their data feeds. Creativity is very hard. Agents are good at providing timely insights, but it is very hard of them to "think outside the box"
  • Autonomy: agents can come up with bad posts, possibly offensive or spammy. Most popular AI agents are currently moderated by humans.
  • Security: It is clear that we are far from AIs that can securely manage money

What's next for the crypto agent economy

Some things I am excited about:

  • Build something agents want. A new wave of protocols and services targeted to agents. Payments (Stripe and Coinbase are rumored to release interfaces for agents payment)
  • Agents for defi: Create index, copy trade, create smart contracts and protocols
  • Creators: share of ai agent content to grow and to form economies around people / agents attention
  • Governance frameworks: Agents that act under mandates and tokens holders vote for their actions. Allowing agents to distribute revenues, manage smart contracts, upgrade codes and so on

Liked this article? Make sure to follow me on Linkedin and X

About me



To view or add a comment, sign in

More articles by Silvio Busonero

Insights from the community

Others also viewed

Explore topics