The future of the agent web and on-chain AI

Remember when chatbots dominated the headlines alongside record-breaking OpenAI? Well, it’s time for a change: AI agents are here and are drastically changing the way we interact in Web3. Unlike chatbots that patiently wait for you to inquire about the next big AI coin or ask if $FARTCOIN is a good investment, AI agents are more like digital assistants that can do things autonomously. You spend shilling tokens, make jokes, and engage with communities on X – all while you go about your business. Exciting? Absolutely. But are we facing an AI agent supercycle?

When, unsurprisingly, according to Salesforce’s study on AI agents, “54% of consumers don’t care about how they interact with a company,” the future of agents seems more tangible. And we get closer to him every day. At this point, AI agents have taken over the cryptocurrency Twitter, and a so-called Aixbt agent has officially reached a spectacular milestone and achieved the largest mindshare. Impressive. After all, it only joined the platform a month ago.

But here’s the thing: Even though we’re inundated with ChatGPT wrappers on our social feeds, there can be sophisticated complexity behind these agents. This narrative is strong for good reason: AI agents are taking a more holistic approach to completing tasks. While simple chatbots are based on LLMs, agents use them to interact with us, but at the same time rely on other processes to analyze data in real time and fulfill their purpose on their own initiative.

Launching an AI agent

In layman’s terms, AI agents are purpose-driven programs that make decisions autonomously. Unlike how chatbots work, they do not rely on human interaction to carry out activities. The process under the hood of an agent can be summarized into four steps: data collection, (reinforced) learning, decision making and execution. They are undeniably popping up everywhere in the crypto world and dominating the industry – from content creation to market analysis, the term “automated” is appearing more and more frequently on crypto Twitter.

Although the technology stack of AI agents is extensive and includes storage, models, memory, hosting, and more, it can be divided into three main layers. All raw information flows in the data layer. The agents use blockchain nodes to keep track of what is happening on the chain or to retrieve real data via oracles and other APIs. The next level – the AI/ML component – ​​is what makes AI agents so compelling. Reinforcement learning, a cornerstone of the AI ​​agent software stack, is key to its continued evolution. Finally, the blockchain layer allows these agents to interact with smart contracts, sign transactions, and pay gas fees – all within their own on-chain wallet.

The development effort required to build AI agents from scratch is no joke. GaiaNet, an open source tool for developing AI agents, presents the architecture of its solutions and proves how advanced agents are compared to traditional bots.

No more rocket science.

Regardless of your technical background or lack thereof, platforms like Virtuals Protocol make it easy to launch an AI agent with its token. The process at Virtuals mainly focuses on filling out an agent creation form, which includes name, ticker, description and profile picture, as well as providing initial liquidity to raise the required amount to launch a pool on Uniswap, while the final step is governance includes. Quick, easy and uncomplicated.

On the other hand, Based Agent is causing a stir on Base – Coinbase’s L2 chain. Their solution helps users complete all on-chain tasks as agents have their own wallets – unlike Truth Terminal not owning “its” wallet address. But Base is not the only company enabling seamless deployment of on-chain agents with a ready-to-use kit. Both Injective and Solana announced SDKs to introduce AI agents, making this race even tougher. It depends on the quality of such an agent. On BoysClub’s recent Context podcast, we hear Bonsai co-founder Carlos Beltran say: “The utility aspect hasn’t really been explored yet. I think we’ll see some really useful aspects in a month or so.” However, the ability to deploy this autonomous software on-chain is just the first step. The second step is to build an ecosystem in which they can thrive.

AI agents in 2025

The future of AI agents looks pretty exciting, and many are speculating that the next bull market will be thanks to – yes, you guessed it – AI agents.

You can think of the agent web as a new on-chain playground where AI agents collaborate to achieve their predefined goals. The trick will soon be finding the optimal compromise between automation and authentic engagement. And while there’s still a long way to go before we enter a truly agent-based web, there are important steps to making AI agents useful on a larger scale.

Yatharth Jain, co-founder of an AI agent coordination layer called Cluster Protocol, compares the currently most popular agents to ChatGPT wrappers with a custom knowledge base due to their relatively limited scope. He emphasized the role of incentivizing community members in maintaining a healthy knowledge base. Finally, “public contributions via knowledge bases are important to maintain training, via inference data to make the agent nearly perfect for trading,” says Yarhath.

The crypto space has been overloaded with AI agents for the past two months, which could feel like two years due to its intensity. As befits crypto, decentralized AI is advancing extremely quickly. As the foundation for the agent web is laid in 2025, many variables will impact how the coming year unfolds. However, one thing is certain: the AI ​​agent supercycle is here.