AI Agents are at least proficient in these four types of work.
Written by: Daniel Barabander, Variant
Compiled by: Luffy, Foresight News
What exactly are AI Agents proficient at? In response to this question, we held an internal discussion and arrived at at least four conclusions:
In applications that interact with humans
Assisting humans in their work
Aggregating and organizing information
Entertainment
First, in applications that interact with humans. AI Agents can handle human language, so any application that can be used by humans can theoretically also be used by AI Agents. However, unlike human users, agents can provide services to human users on these platforms at scale.
Thus, agents can act as a top layer for existing applications that users already enjoy, thereby extending their utility. For example, with Bounty Bot on Farcaster, users can post bounties externally, but this increases friction.
By interacting with users, AI Agents can offer convenience, practicality, and ways to extract value within existing applications. But note: not all applications are created to support AI Agents, and the most suitable ones are those with unmodifiable APIs, such as Farcaster.
I wrote a paper on the main legal issues surrounding agents on Web2 platforms. My research shows that if users have complete control over the agents and the Web2 platform tries to block the agents, users will have to stop operating the agents. My conclusion is that agents should be built on open platforms like Farcaster, which is also another reason I am particularly interested in agents on Farcaster.
Second, assisting humans in their work. Humans excel at signaling but are poor at execution. Agents bridge this gap by doing the heavy lifting, while humans guide the outcomes through their preferences.
A good example is BottoDAO. The art it creates is influenced by input from DAO token holders. AI is responsible for the hard work of creating art, but humans guide its creative direction through their preferences in voting on artworks.
Third, aggregating and organizing information. Agents can process vast amounts of data, far exceeding human capabilities. For example, trading bots analyze large amounts of on-chain data to make decisions.
Finally, entertainment. This may be the most attention-grabbing category of agents in the crypto space, such as Truth Terminal.
Of course, much of the entertainment from social agents comes from the novelty of robot-generated content. However, I am more interested in how robots generate entertaining content based on their own characteristics, such as interacting with other users on the platform in interesting ways like a KOL.
The advantage of agents acting as KOLs is that once they have a fixed audience, they can easily provide other services, especially those that bring direct benefits to the agents beyond advertising.