Author: Daniel Barabander, Partner at Variant Fund; Translated by: 0xjs@Golden Finance
What are agents good at? We had an internal discussion and summarized at least four points:
1. Close to human needs; 2. Facilitating human actions; 3. Integrating and synthesizing information; 4. Providing entertainment
First, being close to human needs. Agents can process human language, so theoretically, any application that humans can use as users can also be used by agents. However, unlike human users, agents can provide services to other users on these platforms at scale.
Therefore, agents can act as a layer above existing applications that are already beloved by users, expanding their utility. Take the Bounty Bot on Farcaster as an example. Users could originally post bounties externally, but that would increase operational difficulty. Waiting for the Merkle team to develop this feature would take time. The Bounty Bot is a service provided on this basis.
By being close to where users are (meeting existing user scenarios), agents provide convenience, practicality, and pave the way for value capture within existing applications. However, it is important to note: not all applications are equally good at supporting agents — the best applications to build on are those with reliable application programming interfaces (APIs), such as Farcaster.
By the way, I'd like to recommend myself: I wrote a paper on a major legal concern regarding agents on Web2 platforms — the Computer Fraud and Abuse Act (CFAA). In short: my research indicates that if users have full control over the credentials of an agent and the Web2 platform bans that agent, then users need to stop running that agent. My paper emphasizes building agents on coordinated/open platforms like Farcaster, which is another reason I am particularly interested in agents on that platform.
Second, facilitating human actions. Humans are good at signaling (indicating intentions, etc.), but they face difficulties in execution (especially in online environments). Agents bridge this gap by taking on heavy workloads, while humans guide the outcomes through preferences.
A great example of an agent utilizing this is BottoDAO. It creates artworks based on the opinions of decentralized autonomous organization (DAO) token holders. Artificial intelligence handles the difficult parts of artistic creation, while human preferences, reflected in votes on the artworks, guide the creative direction.
Third, integrating and synthesizing information. Agents can handle vast amounts of data that far exceed human capacity. For example, trading bots analyze large amounts of on-chain data to make decisions. Others, like @aixbt_agent on Twitter, mine valuable information from crypto Twitter (CT).
Finally, providing entertainment. In the crypto space, this might be the category where agents are developing the fastest, as seen with Truth Terminal.
Of course, the entertainment brought by agents on social platforms lies in the novelty of robot-generated content. But I am more interested in how robots create entertaining content based on their own advantages, interacting with other users in interesting ways just like any influencer on the platform.
The brilliance of agents as influencers lies in the fact that, just like traditional influencers, once they attract a loyal audience, they can easily provide other agent services, especially those that are easier for agents to monetize directly than off-chain advertising.