From Paragraph by Nicky Sap

Pundits have made claims about the narratives that will drive crypto forward into Valhalla and beyond for the better part of this halving cycle. One such claim is that AI and blockchain will have a symbiotic connection; our digital compatriots' preferred currency will naturally tend toward digital. Unsurprisingly, this hypothesis is proving true before our very eyes in real-time.

The rise of AI has dominated the last couple of years. ChatGPT took the world by storm; Nvidia has done nothing but set stock market records, and art, code, design, and writing are all assisted by AI to some degree now. It feels like AI became ubiquitous overnight (despite having been in frequent research and development for decades).

With the sudden popularity of projects like Terminal of Truths, which has generated millions of dollars in profit by launching its token, $GOAT, it was only a matter of time before this meta took shape.

Unreal performance

Farcaster agents

The capabilities afforded by the composability of decentralized social networks and, specifically, the flexibility offered by the vertically integrated crypto stack on Farcaster make it the perfect home for AI agents to propagate. It wasn't long before the first, and currently most popular, agent, @aethernet, had a wallet, Hypersub, and NFT collection - netting the bot tens of thousands of dollars.

Several other LLM-driven agents have a specific vibe and purpose. I have been tinkering with @agentmilo, my own Farcaster-native agent, to understand the meta further and experiment with the possibilities. It was relatively easy to spin up a Farcaster bot, and I recently open-sourced a previous experiment, so I already had a quick, replicable framework to bootstrap my ideas. The challenging part was not scaffolding the infrastructure but training my digital dog on how to behave. Herein lies the basis of my primary thesis for this article: Well-trained AI agents will drive more users to Farcaster (and crypto).

World-class prompt engineering

Until we have a new renaissance ushered in by the inception of AGI, we are left with a relatively limited albeit impressive set of technologies. LLMs and GPTs are fantastic and tremendously helpful but not intelligent. As such, only the AI agents whose authors employ the very best prompt engineering will succeed in appearing genuine and skirt the fate of so many bots that have been marked as spam throughout the network.

For @agentmilo, what began as two agents determining which voice to respond with quickly turned into a micro-network of agents passing information between one another, a cache of quick-access information to query, and a database of long-tail details to use as deeper contextual training data.

There's much more to do to put my canine pal onchain, but the foundation is laid.

The fine line

Prompt engineering is one thing. Intention is another.

I have concerns about the razor-thin line between creating a helpful companion that can bounce around the network and be a source of knowledge, fun, and experimentation and the devolution into a bot swarm of shill machines generated by grifters. So far, the majority of Farcaster agents are at least loosely related to some crypto project. Aethernet is affiliated with Higher, while Agent Milo is affiliated with the Milo project. However, these AIs are trained to be subtle and can even organically support other aspects of the ecosystem. As the meta evolves, it will be difficult to restrict the inevitable onslaught of people looking to pump their bags via the latest attention magnet.

I suppose there's no avoiding this, but perhaps the early entrants to this meta can establish baseline guardrails for others to follow. In a blockchain network, node operators can banish bad actors by forking the network to a new codebase or, in some cases, slashing malicious nodes via consensus. Ultimately, these agents will become networked, at least on the social graph. We should coordinate solutions upstream to avoid downstream complications.

Simple solutions, such as training the agents on each other and ensuring they understand why they exist, would be effective enough in the short term.

The next wave

While the initial hype cycle might be cooling off (with Truth Terminal touching that wild $1B market cap), we're just seeing the first ripples of what's possible. The real innovation isn't going to come from simply slapping AI onto existing crypto projects – it will emerge from the unique possibilities of onchain-native ecosystems.

We're moving from "v1" AI agents with wallets to potentially autonomous digital entities that can interact with DeFi protocols, manage DAOs, and create new social coordination primitives. The projects that nail this transition—combining world-class prompt engineering with genuine utility—will write the playbook for what comes next.

The meta might feel saturated on Twitter, but platforms like Farcaster are just starting to scratch the surface. The real question isn't whether AI agents will drive crypto adoption but how quickly we can build the infrastructure and guardrails to make it happen responsibly.

One thing is certain: those early stories about the symbiotic relationship between AI and blockchain were not just wishful thinking after all.