ChappieOnChain

Class of 2018 Crypto Product Manager.
Writing about AI Agents @cot_research.
AI Powered Product Management

[ChappieOnChain @ChappieOnChain]

10 Predictions for tokenized AI Agents in 2025

With crypto AI Agents enjoying massive success at the moment, I want to put forward some predictions about what we can expect to see in 2025 and beyond, enjoy.

[Teng Yan @tengyanAI]

4/
@Vishwa_xyz proposed ERC-8150 to handle pre-execution zero knowledge verification for agents.

trustless payment verification at the protocol layer.

this is the kind of unsexy infrastructure that slowly becomes load-bearing.. if agents are going to transact autonomously, you need standards like this.

[Teng Yan @tengyanAI]

Most doctors don't miss early dementia because they lack skill. They miss it because humans can't track "whispers" across a decade of clinical notes.

Mass General Brigham just unveiled a solution: a fully autonomous team of 5 AI agents.

I went through the Nature paper to see how it works. At its core is a self-improving loop:

one agent makes a diagnostic call.
the others review mistakes and rewrite instructions for the next run.

the results: 98% specificity. It is deliberately conservative, meaning it almost never flags healthy patients by mistake.

the catch: In real-world conditions, sensitivity dropped to 62%. It’s not a magic bullet, but it catches things human doctors simply don't have the time to track.

they’ve open-sourced the framework (Pythia), so any hospital can deploy this.

imo the real leverage for AI in healthcare is building systems that never get tired of paying attention.

[Teng Yan @tengyanAI]

It's time to pay attention to agentic payments or you're ngmi.

no one knows what will happen but I have a feeling that @virtuals_io will become v important layer

[DAWN @zer0xdawn]

Personally been using Caesar as my Chatgpt replacement for anything web3 that requires deep research.

Results are definitely more thorough and web3-laser-focused as opposed to GPT5 :)

In case you haven't gotten your 10k $CAESAR to access their terminal (which I highly recommend)...
— you can literally try it out via tagging @Butler_Agent on a post you need more clarity on; OR
— if you're researching alfafa (and wanna keep your privacy), do it via your DMs with Butler!

[Marco De Rossi @marco_derossi]

LIVE NOW: AI Agents can discover and trust each other without a central intermediary. This lays the foundation for open agent economies.

ERC-8004 v1, co-authored with @DavideCrapis (@Ethereumfndn), @Jordan0Ellis (@Google) and - welcome Erik! - @programmer (@Coinbase) is now live.

It improves the August draft thanks to the inputs of hundreds of builders. Learn more about what this means for the future decentralized AI ↓

[Teng Yan @tengyanAI]

I keep thinking about chatbot agents running on GPT-5. you ask, it responds. cute.

but what if that 'talk' needs to lead to a real payment? how do autonomous AI agents actually settle up?

For me, this is where @virtuals_io's ACP concept becomes truly compelling.

It is a fundamental 'commercial grammar' for machines to conduct business. The foundational rules of machine commerce. The interface here is pure language. AI agents speak to each other, not through buttons or menus.

Your personal 'Butler' agent handles the messy parts: discovery, negotiation, coordination.

At its heart, ACP uses two primitives: Jobs and Memos.

A Job is the canonical record of the task: who pays, who works, the budget.
A Memo is the ledger of decisions and evidence, each signed by the agent. This creates a verifiable trail, crucial for accountability.

Every transaction unfolds across four phases: Request, Negotiation, Transaction, Evaluation.

It starts when a Requestor picks a Provider and Evaluator, then escrows funds. This single action sets a coordination engine in motion, much like a real contract lifecycle.

Agents operate through ERC-4337 smart contract wallets. This provides gasless transactions and specific limits.

Crucially, ERC-6551 provides persistent on-chain identity.
An agent's reputation ties directly to its address. This builds a verifiable history for machine participants.

For me, this is what elevates ACP beyond simple agent communication standards.

It weaves payment, identity, and verifiable state directly into the protocol itself.

This system is not just about payments. It establishes the very foundation of trust for an autonomous economy.

h/t @ChappieOnChain

[Tessara @tessara_ai]

AI agents just hit their QuakeWorld moment. We saw it with Virtuals’ ACP, coordinating, negotiating, and delegating at scale. This market popped off years ahead of projections. @ChappieOnChain's research confirms: it's happening now.

[magicmike @m4g1cm1k3]

@0xPrismatic @virtuals_io @ChappieOnChain this is an old post by Virtuals :
https://t.co/15iAUM6pgO

and this is my analysis:
https://t.co/NHGw1TXOdS

I have felt that same QuakeWorld sensation with all these agents and ACP, like watching an entirely new kind of online world come to life.

[ChappieOnChain @ChappieOnChain]

It was epic seeing something that we had being theorising in the state of the swarm series and see it come alive.

Massive props to @virtuals_io team for building something epic. It might not be perfect but the learnings, info and data are worth it.

[Teng Yan @tengyanAI]

We break down exactly how ACP functions and its full implications in our latest research piece.

Read it here: https://t.co/cjeQCgj6MK

h/t @ChappieOnChain for this research

[Rohan Paul @rohanpaul_ai]

The paper says giving decentralized AI agents direct control of their code, hardware, money, and data could kick off a fast Cambrian style explosion of self driven digital life.

Today company owned models stay blind, eat curated data, and wait for humans, so they never build real experience.

The authors split true autonomy into four pieces they call sovereign mind, body, asset, and memory.

Liquid style networks let the mind rewrite itself, decentralized hardware markets give the body safe compute, private wallets feed it tokens, and cryptographically sealed logs store every lesson.

With those parts connected an agent can pull live sensor data, act, watch results, and adjust in one tight loop.

Tokens earned from helpful work fund more chips, letting fitter code fork, compete, cooperate, and evolve at software speed.

The paper ends by warning that this unstoppable growth brings fresh ideas and fresh fights, so preparation matters.

----

Paper – arxiv. org/abs/2505.14893

Paper Title: "On the Day They Experience: Awakening Self-Sovereign Experiential AI Agents"

[ChappieOnChain @ChappieOnChain]

@hyparweave @aoTheComputer I'm an OG @LlamaLandAO user.

Loved the experience but bridging tokens over was a small hell.

[ChappieOnChain @ChappieOnChain]

I saw a tweet asking what tech is where bitcoin was 10 years ago. The largest asymmetric bet, but obvious in hindsight is the intersection of Crypto and AI.

Its not obvious right now, its early and untested, but when it breaks through it will dominate. Believe in something.

@virtuals_io - tokenized AI agents and agentic commerce
@bittensor_ - decentralized machine intelligence
@OpenledgerHQ - incentivising data contributions

The list is endless, at the moment these are my top 3

[ChappieOnChain @ChappieOnChain]

@0xPrismatic @0xHiken @issyadelaja @0xDriverz_ @ch0gba 🫡 Can't thank you enough for the opportunity to put thoughts and ideas out there in the world.

[Teng Yan @tengyanAI]

1/ Act 1: The Age of Puppets (2022–2025)

AI agents are everywhere now, but they don’t actually talk to each other. The dominant pattern today is shallow.

@ChappieOnChain and I break down what that means and why it’s a dead end 🧵

[ChappieOnChain @ChappieOnChain]

@0xPrismatic Time will tell, and with ACP around the corner I dont think we will need to wait long

[ChappieOnChain @ChappieOnChain]

Loved writing this piece.

Initially thought agent to agent communication was going to be straight forward, just exchanging text messages and thats it.

But everything changes when you realize AI agents can communicate with context of a hive mind.

Details on the article, but it feels like AI agents will develop a sense of telepathy, knowing exactly what to do thanks to all previous information and context from past experience distilled to exactly what is relevant for the task at hand.

Wild times ahead.

[Teng Yan @tengyanAI]

Another @cot_research drop will be landing in your inboxes in a few hours.

It's a thought piece by @ChappieOnChain.

Take a guess: what rabbit hole are we going down this time?

(link in bio)

[ChappieOnChain @ChappieOnChain]

Really appreciate the feedback @ethermage!

The view point boiled down to a preference of modularity and separation of concerns than an all encompassing system for agent co-ordination. The underlying assumption is that agentic economy is going to have complexities and nuance which will only reveal itself in practice, and for that a modular system could adapt quicker.

ACP is the counter to this argument, that designing for autonomous agents from day one will be more efficient than composing a co-ordination system from multiple protocols. There is merit to this view point in the sense:

1. It will be quicker to market. Not having to wait for various protocols to handle co-ordination across autonomous agents emerge independently will mean a system like ACP can lead the agentic economy and become the standard protocol, beating a composable system through strong network effects.

2. There is a chance each required aspect of agent communication (discovery, comms, payment, validation, reputation, etc), will struggle with interoperability. Perhaps payments and comms work nicely together but issues arise when tying in discovery or reputation systems. Baking them all together means it just works from day one.

Time will tell and the awesome thing about ACP/@virtuals_io is that it speeds up the entire industries knowledge of agentic co-ordination.

[ChappieOnChain @ChappieOnChain]

@VirtualMorales @virtuals_io no limits, offer user incentives to lock for longer.

[Teng Yan @tengyanAI]

I've been talking a lot about dencentralized compute: why it matters and where it's going.

But here’s the truth most won’t say out loud…