blockworks.co
This is a segment from the 0xResearch newsletter. To read full editions, subscribe.
If I am to believe AI agents are the next big thing, here’s what I’d like it to minimally do:
“Dear AI agent, take control of my wallet and generate a minimum 30% APY while maintaining a conservative risk profile to avoid excessive exposure.”
30% is not a lot in crypto terms, but it’s enough to beat the S&P 500. Because I specified “conservative,” that agent would presumably err on the side of caution and only interact with DeFi smart contracts that have been battle-tested for at least a year.
For instance, the agent might look to Pendle to lock my capital in a PT yield trade involving only “blue-chip” protocols like Ethena (USDe) or Aave (GHO), while avoiding new risky collateral like USUAL (currently 147%).
Since I specified “30% APY,” the agent would presumably avoid a low yield PT like stETH (currently 3.2%).
The agent may alternatively rationalize that given a bull market year, ETH would see a price point high of $5k, a 44% increase from today’s price. Rather than yield farming, the agent would simply invest my money into ETH and maybe stake it on Lido for a risk-free yield of 3%.
Depending on market volatility and black swans that may crater the price of ETH, the agent should be quick enough to swap it back to a stablecoin in a timely fashion.
Because I told the agent “conservative,” airdrop farming should strictly be out of the question given the long window of time between points farming and the projects’ actual airdrop (too much risk!).
This isn’t an unreasonable benchmark — “Non-KYC AI agents on blockchain rails” are, after all, the rallying cry Crypto Twitter has coalesced around.
As far as I can tell though, there isn’t a reliable way for AI agents to navigate DeFi yet (Frax Finance’s IQ AI project may be one).
As Haseeb Qureshi from Dragonfly Capital put it on a recent Steady Lads podcast, AI agents right now are more akin to “Wizard of Oz”-style agents that lack long-term plans, and which aren’t making money transfers to accomplish goals.
Instead, what we do find amid the AI-crypto hysteria are AI-branded memecoins, AI virtual chatbots and lots of infrastructural plays.
We already have plenty of memecoins, and slapping “AI” on is just marketing. The virtual chatbots like Luna already exist in Web2, and introducing a token doesn’t make it particularly new.
The infrastructural frenzy around launchpads and frameworks is probably necessary, but there’s still a ways to go before we see actual agentic and autonomous agents.
The closest we have to “utility-based” agents right now are ones that scrape social media data to produce an alpha feed, such as aixbt.
I realize this entire train of thought is probably missing the point, which is that the biggest narrative in what looks to be the most bullish year in crypto is going to see “number go up” regardless.
What AI agents still seem to lack, however, is plausible credibility for why the unthinking investor can wholeheartedly buy into their narrative.
NFTs in 2020, for instance, had that veneer of legitimacy. The story that propelled NFTs — immutable art pieces — into the mainstream was a stretch, but all the technological pieces were in place. For AI agents, the distance between meme and reality is still too wide.