I hope all 9179 of you are having a great first week back so far š„
If you're enjoying my writing, please share Terminally Onchain with your friends in Crypto!
Happy new years everyone! Let's all make 2025 an amazing year. No stress, just momentum š¤
I had a fantastic two weeks off from work and am definitely excited to get back to the research and writing.
Not gonna lie, I just landed back in NYC yesterday and have major jet lag right now. So if this post seems a bit messy, just bear with me till Friday's deep dive. The primary goal of today's post is to restart my brain and get something out. Did I mention momentum??
Although I was technically "on vacation", the truth is that no one in crypto, including me, actually knows how to properly take time off. Especially in a market as crazy as this one.
I was still on Twitter quite a bit. Mainly reading through agent focused articles, following the news on OpenAI's O3 model, seeing the crazy price jumps for agent frameworks, etc.
However, the most important thing I did was bookmark a ton of things that caught my eye! On my flight back, I spent hours going through a bunch of material that helped me form some thoughts I want to share with all of you.
Today's post will be a series of key takeaways that are top of mind for me going into Q1.
Sections below:
Are agent frameworks the new L1s?
The attention rotation to consumer agents
Diversifying trading agents
No adult supervision
Let's dive in.
Are agent frameworks the new L1s?
Over the holidays, the biggest winners by far were agent frameworks such as ai16z, Virtuals, Arc, Griffain, and Zerebro.
ai16z crossed a $2 billion market cap and Virtuals crossed $4 billion! It's worth noting that when I first mentioned these projects in my posts back in late October, ai16z was at less than $80m and Virtuals was hovering at a $350m market cap. If this isn't bull market vibes, then I don't know what is.
Naturally, with prices soaring that fast, agent infra projects have caught the attention of the rest of crypto twitter.
The new narrative I'm seeing is that these agent frameworks are the L1 trades of this cycle. If you were around in 2020-2021, you may remember crypto twitter heavily debating Cardano, Avalanche, Polkadot, etc. And the biggest infra trading wins (by multiples) were easily these smaller cap alt L1s.
Now, in terms of agent frameworks being this cycle's L1 trade, I partially agree and disagree.
I support the analogy as way for people to develop a mental framework on the agent meta. There's definitely a lot of parallels. Virtuals, ai16z, and other tools are the infrastructure layer to support developers building consumer facing agents (see next section). Just as L1s customized their chains to support specific onchain use cases, agent frameworks are specializing to attract particular developer mindshare.
For example, Arc is focused on the small, concentrated set of Rust developers. Virtuals is looking to grow the number of agents working together in its ecosystem. Eliza is directing marketing towards pure open source values and the AI crowd. And ZerePy is the most accessible framework for people looking to get started in Python.
Overall, the comparison to L1s makes sense.
But the reason I don't fully like this analogy is because folks in crypto are obsessed with comparing valuations.
Look, I'm not claiming that ai16z will or will not cross peak L1 valuations from last cycle. My point here is that I'm seeing a bunch of bull posting on the timeline such as "x project hit ___ ATH last cycle therefore...". And I think that's the wrong mentality to have! There are so many different factors to consider and most people aren't in the trenches like the traders tweeting these bull posts. So it creates some kind of false hope or miscalculated price target which might negatively effect people's investment decisions.
My opinion is to keep the L1 analogy strictly to an infra <> consumer analogy. If you want to figure out a target market cap for Virtuals this cycle, then I recommend figuring out your own framework that makes sense for that specific project. For example, what exactly is the TAM in the near term? Is it just the folks playing around in the Solana and Base trenches? Or all of CT? What catalyst gets these agent frameworks to break into tech twitter? All of these things matter so all I'm saying is be cautious of seemingly simple comparisons.
The attention rotation to consumer agents
I made one long form tweet over the break about this section. I personally feel like we're nearing an attention local top with the agent infra frameworks discussed above. Key word there is attention. Not saying price will go up or down. Just discussing mindshare.
There's not much to change from that tweet so I'm going to copy it below. It got a good amount of traction so I'm guessing others are feeling similarly.
(Start of tweet)
Just a gut feeling but I think weāre hitting a local top on agent infra mindshare.
Everyone is now bullish ai16z & Virtuals long term and is holding. After the holidays people will be excited for something new.
My guess is itāll be consumer projects that best demonstrate agentic community management.
Initially youāll see many of the same concepts from 10k pfp projects but the strategies will quickly start evolving as agents try to optimize the quality and quantity of community members.
There will need to be:
an interesting lore and ongoing story
opportunities for fans to engage in meaningful ways
community participation through bounties and proposals. Thinking some version of Nouns style projects for creators to pitch in their own style and taste BUT proposals will be managed by agents. One model could be agents doing a first round pick of top proposals and then a vote with community holders that hold x # of tokens
multiple agents participating in unique ways. This will lead to part of the community rallying around specific agents within one project. What better marketing than friendly competition?
memes, pfps, and sick artwork to share. Also more of a focus on agents tweeting images rather than just text
tiered access to āinfluencingā the lore based on token holding (similar to aixbt terminal)
a store concept to buy merch directly onchain from agents
To be clear, Iām still bullish on infra plays and trading agents, but part of the game is realizing that metas rotate.
People love to be involved and have a say in projects theyāre financially invested in. Thatās why I think agentic community management / NFT style agent projects will catch on quickly. The crypto community is also familiar with these types of projects from last cycle.
Two projects Iām invested in for (nfa dyor) this category are:
I think both teams are executing here, just a matter of time before mindshare catches up imo. There will probably be an āAHAā moment when we see a unique strategy implemented by an agent in terms of community engagement in a couple of weeks to a month from now.
(End of tweet)
x.com/udiWertheimer/ā¦
but youāre not ready for that conversation yet
16
The only thing I'll add here is that other than NFT/IP projects, we'll probably see more attention diverted to agent projects related to gaming and the metaverse such as ArcAgents and Realis. There's a ton of projects in that vertical but I still need to research them so we'll dive deeper in a future post.
Diversifying trading agents
Past the agent frameworks, another token that dominated in December is aixbt, the trading agent launched using Virtuals.
If you've been on crypto twitter at all recently, you've definitely seen a reply from the agent somewhere on the timeline. In fact, it's literally been the #1 twitter account in terms of mindshare in the community even passing posters such as Ansem and Mert.
There are two reasons why the agent is so good:
The developer rxbt has been training the agent on the last 5 years of crypto twitter data so it's learned the community lingo, vibes, etc. I bet if you didn't know aixbt was an agent, you would just think it's some anon schizo trader.
The trading is actually working lmao. It's not life changing returns, but just the fact that this agent is in the trenches and in the green is something significant to me. It's worth noting that most people who attempt to trade anything past holding majors typically end up losing cash. I've even been seeing people copy trade aixbt and are just sticking to it because it works š
Many others have implemented their own versions of crypto trading agents. But I won't go through those, it's clear that aixbt is the winner in that specific vertical.
Rather, what's been catching my attention for a while is the idea of building these trading agents for different asset classes.
For example, one project I've been invested in since it launched is Polytrader agent. It's exactly what it sounds like: an aixbt for Polymarket. It analyzes the open markets on Polymarket, collects and analyzes real time news, forms an opinion, and places a bet. For 500k POLY tokens you can access the terminal, fund a fresh wallet, and customize the parameters for your own agent.
The @polytraderAI agent has been active for about 1.5 days, and I already have 6 positions open, with a total PnL of over 8%
2 positions are winning big (more than 10% return), 1 position is winning small (5% return orā¦
Another project I came across just a few days ago is ARTTO, a trading agent for NFTs and generative art. It's forming taste in art real time and automatically updates its scoring system daily based on its performance.
I don't know how well all of these agents will perform long term but what excites me the most is the idea that there are countless agents still yet to be developed across a wide array of asset classes. They can be as niche as trading Farcaster coins on Base and as broad as trading the entire stock market. The margins will come from what data the agents are trained on and if they can quickly learn from their mistakes and iterate fast enough.
One of Fred Wilson's predictions for 2025 was this:
TikTok turns all videos into memecoins that can be traded on decentralized exchanges all over the world
You can imagine a TikTok trading agent that learns how to best make returns by analyzing what's trending, how TikTok virality works, etc. If you're thinking "ugh this is so cringe why do we need this". Well, your opinion doesn't matter. Because now that this concept is out there, people will do anything to find holes in trading strategies or new asset classes and try to capture the margin. Pandora's box has been opened.
I'll be on the lookout for other agents successfully making bets in different asset classes. One thing I need to do is take time and research how the developers are fine tuning the model to improve the trading strategy. I'm not exactly sure how this works right now but it'll be important to learn because eventually this will be the key differentiator between good vs bad trading companions.
Utility agents
Naturally, as more and more of the crypto community start looking into the agent vertical, there's an increase in doubt and criticism.
I love it. People push back because something catches their eye and makes them care in the first place. Also, it's these criticisms that we should all be focusing on anyways. It can't be price go up and bullish on everything all the time.
One of the best pushbacks I read was from Haseeb (Dragonfly). The actual post is pretty long but I'll share the key takeaways below:
what we're calling agents right now are still glorified chatbots. they're fascinating because the projects are new and crypto twitter needed something fun to get hooked on to
over time, the novelty of these chatbots will diminish and there will naturally be a period where everyone moves onto the next flashy thing until the real agents actually arrive
it's not going to be posting or trading agents that are the 10x use cases in the space but rather crypto software agents
Let's focus on that third bullet point. The idea of software agents is not necessarily new or anything. That discussion is everywhere and we're constantly seeing updates with Claude, Devin, etc.
To me, it seems that Haseeb is specifically talking about agents that can exponentially increase the output of crypto projects and infra.
The screenshot below is a bit wordy, but take a second to read it:
In the first bullet point, the part that caught my attention was "self-financed projects like Hyperliquid and Jupiter will go from the exception to the norm". I've been saying something similar for the last year now. Not necessarily because of agents or anything, but just the idea that tokens and protocol incentives make it a lot more realistic for solo devs to bootstrap their businesses. But the idea of crypto software agents makes the idea way more robust. One of the biggest discussion points in the space is how there's a lack of consumer projects. With the proper tooling, hopefully a lot more folks are able to pitch in.
This thread by 0xdesigner is relevant here. Core point being that he wanted to build an app as a designer and tried his best using current AI tooling but it was a lot harder than he expected. With actual agents that can take tasks from end to end, I'm guessing the experience will be much more different.
it was the happiest day of my life.
here's why i don't recommend anyone doing the same.
my learnings and reflections:
The second bullet point in the screenshot was something I hadn't even thought of but makes so much sense. Agents focused on crypto security (that actually work) might be one of the highest +EV projects in the space. 24/7 security, patching ability, monitoring systems, etc. would totally change the narrative on how crypto is perceived.
Still tons of research to do here but two quick examples:
H4CK Terminal - first ever white-hat AI agent hunting vulnerabilities, securing funds & redistributing bounties
Soleng - World's First Solutions Engineering and Developer Relations Agent for Web3
Excited to see more development here. Would be a huge bump for the agent vertical if tier 1 devs in the Ethereum and Solana space started actively using these agents.
No adult supervision
Last takeaway I want to share is that it's clear the verifiability of agents is becoming increasingly important.
Up until now, the novelty of crypto agents was leading the excitement. But as we start to see the market get saturated in the coming months, many will start focusing on how "agentic" these projects actually are.
Remember, the core point of an agent is that it handles tasks end to end without any supervision. Many of the projects we see today are not fully there, the dev still leads the way.
For true economic agency, agents must be able to autonomously manage their own money. This can start to evolve agent behavior since you can place an economic constraint on an agent so that it must pay for its own inference costs. This introduces Darwinism to agents so that they must earn revenue in order to survive. - @0x3van
This is where TEEs (trusted execution environments), proof of sentience, secure storage, etc. all come into play.
L2 beat focuses on what stage of decentralization rollups are at.
similarly, agent beat will verify how independent agents actually are.
Are they self-operated like @freysa_ai? How much of the crypto x ai stack is being used?ā¦
We'll dive more into TEE and projects like @galadriel_ai to understand what an "agent beat" would even look like. Interesting opportunity in my opinion...maybe someone from the terminally onchain community wants to build this?
That's all for today's post!
I ended up having a lot more on my mind than I thought.
Let me know if you think there are other takeaways I may have missed or you've been noticing.
I hope all of you have a great rest of the week and I'll see you on Friday!
- YB
