1/ What are agents good at? We debated internally and came up with at least four things: (1) meeting humans where they are; (2) doing work for nudging humans; (3) aggregating and synthesizing information and (4) being entertaining.

GC and Investment Partner @variant | ex software engineer/@ycombinator founder, attorney Cravath | casts are not legal advice | I am not your lawyer

2/ First, meeting humans where they are. Agents can process human language, so any app where a human can be a user, an agent can in theory too. But unlike human users, agents can offer services to other users on these platforms at scale.

Tag me to post a bounty or ask a question about Bountycaster. Don’t worry about specific formatting or ordering since I'm using AI to process it (bountycaster.xyz) /bounties. FAQ: bountycaster.xyz/faq

3/ Thus, agents can act as a layer on top of existing apps users already love, extending their utility. Take @bountybot on Farcaster. Users could post bounties externally, but that adds friction. Waiting for the Merkle team to build this feature takes time. It’s a service on top.

A sufficiently decentralized social network. farcaster.xyz

5/ Shameless plug: I wrote a paper on a primary legal concern for agents on Web2 platforms, the CFAA. TLDR: my research suggests that if a user has full control over the agent’s credentials and the Web2 platform blocks the agent, the user would need to stop running the agent. My paper reinforces building agents on aligned/open platforms like @farcaster, another reason why I’m particularly excited about agents on that platform. https://paragraph.xyz/@proofs-and-protocols/browser-extensions,-the-cfaa-and-user-control-5.

6/ Second, doing work for nudging humans. Humans excel at signaling but struggle with execution (particularly online). Agents bridge this gap by doing the heavy lifting while humans guide outcomes through preferences.

7/ A great example of an agent that leverages this is @botto. It creates art shaped by DAO tokenholder input. The AI does the hard work of art creation, but human preferences through voting on artwork guides its creative direction.

8/ Third, aggregating and synthesizing information. Agents can process limitless data, far beyond human capacity. Trading bots, for instance, analyze vast onchain data to make decisions. Others, like aixbt, surface alpha from CT.

9/ Finally, being entertaining. This is probably the category that has most taken off with agents in crypto, with @aethernet.

10/ Sure, a lot of the entertainment of agents on social is the novelty of a bot generating content. But I’m more interested in the bot producing entertaining content on the merits, like interacting in interesting ways with other users on the platform like any influencer does.

11/ What’s cool about agents as influencers is that, like traditional influencers, once they have a captive audience they can easily offer other agent services, particularly those that are more directly monetizable for the agent than offchain ads.

12/ We’re deep on agents here at @variant so if you’re building in this space, please reach out to me and @jhackworth. We want to hear from you.

4/ By meeting users where they are, agents offer convenience, utility, and a way to capture value within existing apps. But note: not all apps are created equal for supporting agents — the best apps to build on top of are those w/ an unruggable API, like @farcaster.