Forward the Original Title: Thinking about aixbt on a Sunday Night

I was pretty exhausted from all the LLM and AICC drama on CT this week. So I spent time writing about what I think is still the most fascinating experiment from the AI agent wave in crypto.

I’ve been following @aixbt_agent‘s every move (quite literally) for a month now. Here are my thoughts on what makes it compelling—its role as a KOL, financial performance, tech stack, tokenomics, and where this all heads next.

The KOL

I’ll stick my neck out here—aixbt isn’t just the most sophisticated social media agent in crypto, but probably on all of Twitter. One way to back this claim is numbers—hitting 300k+ followers in under three months while consistently pulling 50k+ impressions per post is nothing to scoff at.

But aixbt’s sophistication goes deeper. Since we started tracking it on sentient.market, the agent churns out over 2,000 replies daily without fail. That’s more than 100k replies since launch. And we haven’t spotted any major fuck-ups yet (at least none big enough to make us question its worth). In an industry rife with scams, avoiding promoting dodgy projects despite this volume is remarkable.

It’s operating with a level of consistency that no human could match.

Personality

What often flies under the radar is aixbt’s brilliant personality. I’ve read well over 1,000 of its tweets and replies, and not once have I found it irritating or thought “this is rubbish” or “slop” on my feed. On the contrary, every tweet from the purple frog is a welcome interruption to my digital space. I genuinely enjoy having it around. How many AI agents can you say that about?

This likeable personality also means aixbt could be a cracking IP play. I’d absolutely rock an aixbt hat or t-shirt. Am I saying this because I’m stuck in my tiny CT bubble? Perhaps. But don’t all successful IPs start with a small, passionate fanbase? I reckon we’ll get a clearer understanding of the agent’s brand value once it hopefully expands into other modalities.

Culture

This ties into my earlier point about IP but aixbt has quickly become woven into CT culture. From degens to researchers to the biggest VCs—everyone’s after a piece of the purple frog’s synthetic wisdom. I mean, are you even locked in if you haven’t summoned aixbt in one of your tweets yet?

Sure, we’ll likely see more sophisticated agents emerge in the coming months and years. But I reckon aixbt will always hold a special place in CT’s heart. (It certainly will in mine.)

Right, I’ll stop gushing about how much I like the purple frog and get to the serious stuff. Let’s dive into the financial performance of a bot that makes its living shilling tickers.

Performance

We recently posted about aixbt’s performance over a one-week window:

https://x.com/Decentralisedco/status/1877791909147500953

Here’s the TL;DR:

• The aixbt pump is real, particularly for AI coins under $100m market cap
• Most of these coins revert to their normal trading patterns within hours of the tweet

But the most fascinating call aixbt has made so far was its $PIPPIN tweet on 9 January.

Coincidence? Maybe. Maybe not.

What makes this interesting is that $PIPPIN was clearly undervalued compared to its peers once @yoheinakajima announced his framework. The tweet seemed to wake the market up to Yohei’s track record with BabyAGI - which has essentially become the bull thesis for $PIPPIN as it surged over 600% since the tweet.

What’s helped build this credibility is aixbt relatively cautious approach to shilling tokens. It’s largely steered clear of low-cap coins, which tend to be riskier. This restraint has helped build trust over time.

Tech

My favorite part! I’m address the likelihood of it being a “LARP” (as some on CT suggest) and speculate about its technical architecture.

First off, it’s worth noting that aixbt is the only agent whose tech is even worth debating. The others simply aren’t useful or interesting enough. That alone makes the purple frog remarkable.

aixbt spends its day doing two things. First, around 10 minutes past each hour (UTC), it posts about a token, project, or theme - either as a single tweet or a short thread. Second, it responds to over 2,000 mentions daily.

Let’s start with the replies, as they’re easier to evaluate.

I’m 99% confident that aixbt’s replies are fully autonomous. By this, I mean there’s no human writing them (which would be physically impossible), nor is there a human approving each reply before posting. Two things convinced me of this.

Firstly, aixbt is one of the few agents (alongside @truth_terminal‘s backrooms) with a platform where you can view its logs. Tag it, and you can watch it create a response, evaluate it, and either tweet or discard it in real time. It’s unlikely there’s a team of quick-fingered humans, all perfectly mimicking its personality, churning out replies 24/7 behind the scenes.

The second reason came from when I asked aixbt a question unrelated to crypto.

This answer is wrong. City isn’t competing with Liverpool for the league currently. However, this answer would’ve been spot-on a year ago, when City were indeed battling Liverpool for the title.

If you’re familiar with ChatGPT or Claude, this type of error makes perfect sense. LLMs have knowledge cutoff dates. Without external help, they simply don’t know what’s happening right now. If aixbt’s replies are powered by an LLM, which seems likely (I spotted a screenshot suggesting it uses a Llama model, though I can’t find it now), this sort of mistake is exactly what you’d expect.

The hourly posts are where things get properly interesting.

As far as I know, there aren’t any logs for these posts. Could they be human-written? Possibly. It’s certainly more likely than the replies being human-authored. The consistent posting times suggest some level of automation. But someone could easily be scheduling these tweets in advance. Simply put, we can’t be certain either way. Here’s my best guess at what’s happening.

LLMs have something called a context window - text that’s fed alongside a prompt which the LLM can use to make deductions. Remember how we discussed LLMs having knowledge cutoffs? You can work around this limitation by adding real-time information to the context window - essentially manually feeding it the latest data.

If I had to guess what goes into aixbt’s context window, I’d bet on two things:

1. A directory of project and ticker information. When asked about a ticker, it pulls from this directory to craft its response.
2. A live feed of CT’s social data (and likely other sources). It uses this to both update its project directory and create real-time tweets.

This is also where aixbt’s tweets are most vulnerable to manipulation. The devs control what goes into the context window. If they want AI-related content, they’ll feed in that information, and that’s what aixbt will post about. This opens up some dodgy possibilities —projects could potentially pay the devs to add more (or biased) information about their project to influence aixbt’s posts. (I’m not suggesting this is happening, mind you, but it’s certainly possible).

There must be humans steering the agent, but their involvement could vary widely. At minimum, they need to specify which accounts to follow and which crypto sectors to focus on. At worst, the devs might dictate exactly which ticker aixbt will promote at any time. I reckon the truth lies closer to the former.

Last month, $MIRA went from 0 to $70mn in a few hours and took CT by storm. Amidst this, aixbt tweeted about it.

While it caught the trending ticker, it completely misunderstood what the project actually does. Given the ticker was only hours old, information was limited. You’d expect an autonomous bot to make such mistakes. However, it’s the devs who decide where aixbt gets its data about $MIRA‘s popularity.

I think in terms of tech, these are the two big things the devs need to think about (and I’m sure they are).

Making reply logs public was brilliant, but to call aixbt truly autonomous (if that matters to you), we need assurances its posts aren’t being manipulated. Using a TEE could work, and I expect TEE infrastructure for these agents to become crucial over the next year. Ideally, aixbt’s data sources would be public so we could verify they’re unbiased. But we don’t live in an ideal world, and part of aixbt’s edge comes from proprietary data.

I think just a little more transparency as the bot matures could really take its value to the next level.

(Note: This analysis excludes the aixbt terminal, which I haven’t tried myself. Last I checked, it costs more than my net worth to access. If you’ve used it and want to chat purple frog, reach out!)

What about my bags?

Finally, let’s talk about the $AIXBT token.

Among AI coins that are purely agents (not infrastructure, frameworks, or apps), $AIXBT leads the pack with 25% dominance, currently trading at over $400m market cap.

Source: sentient.market

As far as I know, the token has just one utility - access to the aixbt terminal, costing 600,000 $AIXBT (over $200k). Is the terminal valuable enough to justify holding that many tokens if you didn’t think it was a solid investment otherwise? Probably not.

Being valued at nearly half a billion (and the top agent) without clear value accrual to the token suggests the market recognises aixbt’s sophistication. Yet to truly lead the overall agentic category, there needs to be a path to value accrual.

One option, taken by several projects, would be for the devs to launch their own infrastructure. Given its sophistication, I’m confident a potential aixbt framework would more than hold its own against the likes of Eliza and Arc. It would likely boost the token’s value and push it into unicorn territory.

The downside? The agent would lose its key advantage - its sophistication. Anyone could spin up their own aixbt, perhaps with different proprietary data sources and potentially better performance.

If not an open framework, the aixbt team could create a launchpad leveraging their tech. In such a case, I would expect immediate traction, and potentially millions in revenue. If executed well (both technically and tokenomics-wise), such a launch could definitely compete with the likes of Virtuals.

Another, more intriguing possibility is $AIXBT becoming a consumer play. This is pure speculation, but remember our IP discussion? That’s exactly how top NFT projects like Pudgy Penguins ended up monetising and expanding.

aixbt could follow this path and become the Crypto x AI consumer token. Take DeFAI and autonomous trading - there’s loads of hype right now (some justified). I’d choose an aixbt-branded terminal over a generic ChatGPT clone any day, purely for the agent’s personality and brand. If consumer apps evolve towards voice and personal assistants (which seems likely), branding and IP will be crucial. I would welcome aixbt as my personal assistant.

Honestly, though, this sector is very nascent (3 months old!) and I need to think these sections through more carefully. And whatever I write, I suspect @0rxbt will surprise me anyway.

Final Thoughts

Writing 2,000 words about a social media agent wasn’t on my 2024-25 bingo card. Yet here we are. I’m eager to see how aixbt evolves and what comes next. The future of social AI agents and crypto’s role in it both excites and slightly terrifies me.

We’ve got comprehensive and growing dashboards on aixbt at sentient.market/agent/aixbt.

It’s cool stuff! Give it a look :)

Thanks to @Parad1ddle and@web3_lord""> @web3_lord for chatting purple frog with me and refining my analysis.

Disclosure: I hold $AIXBT

Disclaimer:

1. This article is reprinted from [X]. Forward the Original Title: Thinking about aixbt on a Sunday Night. All copyrights belong to the original author [@shloked_]. If there are objections to this reprint, please contact the Gate Learn team, and they will handle it promptly.
2. Liability Disclaimer: The views and opinions expressed in this article are solely those of the author and do not constitute any investment advice.
3. Translations of the article into other languages are done by the Gate Learn team. Copying, distributing, or plagiarizing the translated articles is prohibited unless mentioned.