DappRadar is not only building a platform to explore and discover decentralized applications, but also a decentralized intelligence system to supercharge the way we search information in Web3. At ETH Bucharest, I lifted the veil on what the team behind the World’s Dapp Store is building next: Next generation discovery tools powered by artificial intelligence.
Since its inception in 2018, DappRadar is the leading platform for dapp discovery. Often dubbed the World’s Dapp Store, the platform allows users to discover and explore decentralized applications and blockchain ecosystems through verifiable data. Currently DappRadar supports over 85 different blockchains and close to 18,000 applications.
Blockchain data comes with an information overload and terabytes of data across all these blockchains. Moreover, additional information comes from X, Discord, Telegram, blog posts, forums, analytics tools, and the list goes on. Moreover, Web3 operates 24/7, new projects launch every day, and the entire industry runs without breaks, without timezones, and without holidays. The landscape keeps evolving rapidly, and new trends and narratives emerge daily.
It doesn’t come as a surprise, but humans struggle to keep up with Web3.
Web3 was built for machines
On DappRadar we don’t speak about daily active users, but about Unique Active Wallets, abbreviated as UAW. Many of these wallets aren’t controlled by people, but by bots and AI-powered systems. The blockchain space is a machine-native environment, and no matter how hard we work as an individual, we can’t keep up.
The solution is not to work harder, but to work smarter. We need digital delegates to represent us in this machine-native environment. This is where AI agents come into play.
AI agents as your tireless Web3 delegates
You can’t gather and filter all the information coming from the blockchain industry, but AI agents can. Just like the blockchain, AI agents never sleep. And contrary to you and me, these AI agents also never get a burn out. These agents read, write and speak code, just like everything on the blockchain.
Through code, we can give the AI agents certain rules to live by. This will then allow the AI agents to make autonomous decisions, and execute onchain actions without delay. The agents can do this by plowing through gigabytes of code and complex data per day. Moreover, AI agents excel at these types of tasks.
Within Web3, AI agents can help you and me in many ways:
- Monitor the market by tracking projects, blockchains, following price movements and measurement community and KOL sentiment.
- Summarize key points from governance proposals to help you in your decision making process without the need to read entire whitepapers.
- Manage your portfolio and send alerts about risks or opportunities based on your token holdings.
- Execute transactions based on pre-set parameters, for example initiated by a combination of onchain activity, token price action and social sentiment.
Building vertical AI agent swarms
With the team at DappRadar we are building vertical AI agent swarms. Instead of one general purpose AI for all of Web3, we’re building specialized teams of AI agents. These teams, or swarms, focus on one particular project. Agents look at blockchain data, blog posts, social media activity and so on.
Project swarms exist of multiple agents, each responsible for a different task. For example, one project swarm can have 5 agents:
- Scan social media like X and Farcaster to get an idea about community sentiment
- Analyse governance proposals and voting outcomes
- Follow community conversations on Discord and Telegram
- Get all the technical information from whitepapers and technical guides
- Gather and analyze onchain data, such as transactions, tokens and NFTs.
In addition, on top of these project swarms we envision ecosystem or category swarms, which specialize in blockchain sectors like Gaming or DeFi, or focus on one particular ecosystem, such as Arbitrum or Ethereum. These swarms can combine information from multiple project swarms and their own general information. This can be useful to for example find the best yield opportunities in DeFi or earning potential in play-to-earn games.
At the top of the pyramid, there’s an all-controlling master agent. This agent connects all swarms, and has the ability to delegate tasks. You could say that users always interact with this master agent, which then delegates the tasks or questions to agents and swarms down in the pyramid.
The challenge of trust
The rise of ChatGPT and other AI services has paved the way for a new paradigm in the way we access information. However, AI agents aren’t flawless yet. In order for AI to find real adoption in Web3, various critical obstacles need to be overcome.
First of all, AI can present false information as facts. An AI can report incorrect metrics, and simply make up features and historical data. In Web3 these hallucinations can cause financial damage, because the AI may provide false trading signals or it may misinterpret smart contract risks. Just two months ago an AI agent transferred $100,000 worth of ETH after it was tricked in doing do.
In November last year users managed to override the one and only directive of an AI agent, to not transfer any money. As part of a challenge, users needed to pay to send messages to the AI agent, increasing the prize pool. In the end one user managed to convince the agent to send $50,000 to their wallet.
Finally, new technology such as AI agents comes with its own set of security concerns. For example, AI agents may be manipulated by crawlers, or base their knowledge on scam sources. Moreover, these agents might miss critical security information, or bias towards high volume, but low quality data. In the example below, the AI agent agreed to selling a new car for $1 after chatting with a user.
Obviously these types of scenarios are an issue, and we can’t trust AI agents simply by their word. AI agents need to follow rules, provide clear information, and should never lie. But, how do we achieve this level of trust?
Building trust
In order to build trust, we must verify the information AI agents share and the source they use. This is where DappRadar’s long awaited contribute-to-earn program comes into play. We envision a human-in-the-loop system where the community watches over the AI agents.
These agents analyze data, find patterns and suggest actions. But then human curators need to validate the information and the suggested decisions. In order to make that possible, the community needs to rate information and data sources, while also validating the output by the AI agent.
The result will be a feedback loop which enhances the information accuracy of AI agents over time. The result will be a reputation framework where the accuracy of every AI agent is tracked and data sources get rated by the community. Moreover, through contributor rewards, the community has an incentive to curate and verify to the best of their ability. In the end we will have a trust system where transparency and reliability are key.
Closing words
The collaboration between AI and Web3 is inevitable. DappRadar is keen on revolutionizing the way we discover applications, powered by AI and blockchain technology. Our agent swarms will serve both users and builders alike, while maintaining high quality data. With contributions from both the community and agents, we’re building a decentralized future that’s accessible and trustworthy for everyone. With our approach we address the challenges, while maximizing the potential benefits for all participants.
