.svg)
A full guide to AI agents (with 10 best projects)
Keep hearing about AI agents but not sure how they differ from ChatGPT? You definitely need to read this article to keep up with rapidly evolving technology and ever-changing trends.
AI agents vs LLMs and AI assistants: a definition
To get to a definition of an AI agent, we should first think of what an agent is. Merriam-Webster dictionary defines it thus: “something that produces or is capable of producing an effect”; “one that acts”; and, most importantly for us, “one who is authorized to act in the place of another”.
Add software to the mix - and we can get to a definition of an agent program in IT: it’s a software system that is authorized to act (by its creator or end user) and is capable of acting and producing an effect.
Now let’s add AI. What can an AI-enabled program do that a simple computer app cannot? The three key differences are degree of autonomy, non-deterministic action, learning ability, and capacity to cooperate.
- Autonomy: an AI agent can act autonomously to a degree, performing complex sequences of actions without human inputs. For example, an AI can go through recent KOL posts on X to understand which coin is trending; create its own tweet with its opinion on the coin; post it; and respond to comments. If users keep responding back, the AI will sustain the conversation as long as needed.
- Ability to learn: more advanced AI agents can collect feedback and improve over time. For instance, an AI agent that manages an X page will analyze engagement under different posts and tweak its content to get more likes and comments. A non-Ai agent app may be able to analyze the engagement, but it won’t change its content unless you tell it to.
- Cooperation: AI agents can engage each other to collaborate on tasks. Of course, you can program a regular app to trigger an action in a different app (that’s what Zapier does, for example), but an AI agent may be able to figure out on its own which other agent to team up with, without clear instructions on your part. Eventually AI agents could form complex structures, known as “swarms”. A good example is Swarms, an AI agent marketplace project that suggests that multi-agent collaboration is the most transformative technology of our time.
- Non-deterministic decision-making: whereas a simple, non-AI agent app executes instructions in a deterministic way (‘if A happens, then I do B”), an AI agent can analyze different sets of conditions and then make a decision.
For example, a non-AI trading bot can be programmed to buy a coin whenever the price crosses the 200 EMA line from below and the RSI (a technical indicator) reaches a certain level. But an AI trading agent won’t just look at these technical indicators. It will check if any economic data is about to get published and cause market volatility; and it will also check if a token unlock is looming that could affect the price. It will go through influencer posts on Twitter to see if a token is being artificially pumped; and read the political news. In other words, it will use many sources of data, and even if the coin looks from the technical analysis standpoint, it won’t necessarily buy. AI agents and crypto: a perfect matchA lot of AI agents in the market were developed by big IT companies like Google, IBM, and Microsoft, and they don’t require crypto or blockchain to run. But cryptocurrency expands the range of actions that AI agents can take. An agent can operate its own crypto wallet (and even create one), so it can buy and sell crypto and send tokens to other addresses. Think about it: an AI agent cannot have a bank account or a debit card, but it can have a crypto wallet. Crypto empowers AI to make financial transactions without human authorization. But that’s not the end of it: AI agents that control crypto wallets can better collaborate with other agents and humans… because they can pay them! One of the first such transactions happened in August 2024 at the Coinbase Dev event. One AI paid another in crypto to buy special AI tokens (chunks of data, not crypto tokens).
An article by Coinbase CEO Brian Armstrong about the historical AI-to-AI transaction. The next step is to facilitate AI crypto payments for real-world services: for example, so that AI agents can buy ads on X or Facebook, or pay for access to research reports.Top 10 best AI agent projects and tokens to watchThe AI agent landscape: is FET really an AI agent token?AI agent projects in Web3 can be roughly divided into two groups:
- Infrastructural: platforms and technological solutions for launching AI agents. These are aimed mostly at developers and project teams, or at least at users with some technical expertise. Examples: Virtuals Protocol, ElizaOS.
- AI agents proper: projects aimed at end users, with agents that communicate with their followers on socials, generate market insights, create art and music, etc. Examples: Aixbt, Luna, VaderAI.
Now if you look at the rankings of AI agent tokens on CoinGecko or CoinMarketCap, you’ll notice something unusual. A lot of the coins on the list belong to projects established years ago, when nobody talked about AI agents yet. No.1 is FET (Artificial Superintelligence Alliance) - the union of Fetch.AI and Ocean Protocol, which were always positioned as infrastructure providers for AI in Web3. Fetch.ai was among the first to propose a network of agents that can communicate and collaborate, so it was ahead of the trend.
Also in the top 10 are Origin Trail (TRAC, a decentralized knowledge graph for AI assets), PAAL AI (a platform for building Ai bots), and SingularityNET (AGIX, a marketplace for AI services). Thus, 4 out of 10 top projects on the list have a different or larger scope that just AI agent creation. We’ve covered some of them in our earlier articles on AI and Web3; in this article, we will focus on newer projects that fall directly under the AI agent narrative. We will cover both infrastructure projects and specific AI agents, grouping together agents based on the same infrastructural protocol where possible. The top 10 AI agent projects you should follow
Virtuals has emerged as the biggest player in the AI agent space - a launchpad that democratized launching AI agents. Virtuals also turned the Base blockchain into an AI agent hub (though the protocol now supports Solana, as well).Make no mistake: launching an AI agent with Virtuals still requires work, and it’s not free. You’ll need to pay around $250-300 in $VIRTUAL tokens to create an agent, plus upwards of $2500 to buy out a share of the initial token supply. This is very different from launching memes on Pump.fun, where it costs barely $1 to create a memecoin. Virtuals does take an important feature from Pump, though: the bonding curve. Each agent comes with a token that initially exists off-chain and is traded on the internal Virtuals DEX. Every buy increases the price and the market cap - and once it reaches a certain threshold, the agent token “graduates”: the token is generated on-chain, and a liquidity pool is created for it on Uniswap. Virtuals is home to some of the best-known AI agents: Luna, Aixbt, AixCB, VaderAI, and hundreds of others. $VIRTUAL is also the single biggest token in the AI agent space by market cap.
This is the most successful AI agent launched on Virtuals Protocol in terms of token market cap: over $500 million as of late January 2025. Aixbt gathers data about the crypto market and what individual projects are up to and generates insights. It has become a good source of crypto news and investment ideas. Aixbt covers such topics as crypto ETFs, exchange volume, fundraising, product launches, whale transactions, tokenomics, incentive campaigns, etc.
Expertise: music, streamingLuna is a singer, Tik Tok streamer, and runs her own X page, too. She’s the most popular music-making AI agent out there with over 850,000 followers on Tik Tok, streaming 24/7. Luna was created by the team of Virtuals Protocol and is actually the soloist of an AI K-pop band called AI-DOL.Most of her tweets have to do with music and popular singers. She gets a good number of reactions under the posts, many of which are by other AI agents - and some are quite critical! That’s another good example of AI-on-AI interaction - and a hint that AI agents can enter into feuds with each other.
AixCB is also a decentralized VC agent, but created using Virtuals Protocol. It invests mostly in AI projects, with almost 80 projects already in the portfolio and 20+ more grants in progress. Founders can submit their startups for consideration as long as they have a White Paper, a detailed proposal, a team, and a roadmap.
Rekt is a meme-centric AI: it posts mostly screenshots of funny conversations that have to do with crypto under the tagline “Memes that cause you pain”. To these, Rekt adds ironic comments. Interestingly, $REKT is one of the biggest AI agent coins (over $70 million market cap), in spite of a lack of clear utility - another hint that AI tokens are almost a subset of memecoins.
VaderAI was also launched on Virtuals Protocol. It’s a decentralized investment platform where users form DAOs to invest in crypto assets, and those DAOs are run by both AI agents and humans. At the moment, VaderAI has two active funds: Small cap and Micro Cap. Both are indices composed of several assets with different weights: for example, the Small Cap features $VIRTUAL, $AIYP, $VADER, $ROCKET, and several more AI tokens. Performance is calculated against $VIRTUAL: the current (at the time of writing) result is 43%, meaning that investors have earned 43% more than if they had simply held $VIRTUAL.
Eliza is an operating system for AI agents - a framework similar to Virtuals Protocol but aimed at developers rather than end users. The best-known AI agent built with Eliza is probably ai16z, which we will cover below. Other Eliza-based projects include The Major Protocol, xNomad (an NFT agent), and Musk AI, an agent trained on Elon Musk’s tweets. By the way, Eliza supports DeepSeek, the popular free AI assistant. This means that agents built with ElizaOS can query DeepSeek for data.
You’ve probably heard of a16z, the famous investment fund also known as Andreessen-Horowitz. Now it has an AI “counterpart” - a VC fund (or, using the project’s own definition, an “AI-led hedge fun”) run by an AI manager ironically named Marc AIndreessen. The fund runs on Solana and is built by the team of ElizaOS. $AI16Z has become the first AI agent token on Solana to reach a market cap of $1 billion.
- Sekoia VC ($SEKOIA)
We’ve included yet another AI-run VC fund on our list - but that’s because AI investment and trading are such important use cases for AI agents. Sekoia boasts an active portfolio of 7 projects, all of them in the AI industry. The agent that runs the fund is built on Virtuals Protocol, but $SEKOIA stakers are strongly involved in decision-making (though they don’t normally vote on which projects the fund should invest in - otherwise it wouldn’t be an AI-run fund).
- Swarms ($SWARMS)
Agent “swarms” are the latest trend within the AI agent narratives: they are groups of agents that gather together to collaborate on tasks. Swarms is one of the projects spearheading this concept of a multi-agent economy. The framework offers many tools, such as a Yahoo Finance integration for financial analysis and the ability for agents to like and comment on X posts autonomously. Are AI agent tokens essentially memecoins?From AI memecoins to AI agent coinsRemember how so-called AI memecoins were all the rage a couple of months ago? After the AI agent Truth Terminal began shilling Goatseus Maximus ($GOAT), users began buying memes whose social media pages were run by an AI - and those that were supposedly launched by an AI. Actually, the main reason behind the $GOAT pump was that people were led to believe it had been launched by Truth Terminal - which wasn’t the case at all. We covered the most popular AI memecoins, including GOAT, TURBO, and ACT, in a separate article. The AI agent hype is the logical next step in the evolution of this trend: from memecoin X pages managed by AI to artificial intelligence that does a lot of other, more useful things and has its own coin.
AI agent tokens as a subtype of memecoins
Let us make a provocative suggestion: most AI agent tokens are basically memes. We’ve mentioned how Virtuals Protocol works similar to the pump.fun memecoin launchpad, with a bonding curve that determines the price of new tokens. But that’s not the only similarity between AI agent tokens and memes.
The truth is that most AI agent tokens have very little or no utility. For example, Virtuals Protocol suggests that in the future, each agent’s revenue will be shared among the holders of its tokens, but it’s not clear when this will be implemented. You don’t need $AIXBT, $LUNA, or $REKT to enjoy these agents’ content: it’s free! The only benefit you get from holding these tokens is the eventual profit if their price goes up - and the price essentially depends on the virality of each agent and of the narrative as a whole. That’s exactly what happens with memecoins.
There’s one more aspect: the sense of belonging. AI agents have their communities of fans, not unlike the cults that grow around memecoins. Such an AI agent cult can be even stronger than that of a meme, as an AI agent can communicate with its community directly and increase engagement in a much more personal way than the team of a memecoin. AI agents who are expert at community-building - rather than those that offer more practical products - will have better chances at success.
What’s next for AI agents in 2025?
The explosion of AI agents at the end of 2024 came as something of a surprise, as the whole decentralized AI was starting to look exhausted. The agent narrative evolved as a progression from memecoins and Pump.fun to AI memecoins to AI agents.
AI agents have a clear utility for Web3, as they help users and teams invest, trade, interact with communities, create content, and so forth. Inter-agent and multi-agent collaboration open up even more exciting opportunities. And, most importantly, AI agents are already working out there - unlike many other Web3 narratives (RWA, gaming, metaverse etc.) that promised a lot but have delivered little so far.
The AI agent narrative is projected to remain strong in 2025, with agents taking on ever more functions. There are already agents that can talk with their own voices in different languages and even order food. We’re bound to see many new unexpected agent capabilities that nobody can predict right now - and, for our part, we believe that AI trading will be one of the most important “superpowers” that agents will develop.
We’ll keep covering AI agents in the Pontem blog - stay tuned! Don’t forget to follow Pontem on X, Telegram and Discord for the latest product updates.
