Jarrod Watts Profile picture
Dec 10 16 tweets 5 min read Read on X
AI Agents have now reached over $10 billion in market cap.

So, are AI agents the next big thing, or just another way to pump tokens?

To find out, I spent the past few weeks learning about how AI agents work under the hood.

Here’s what I learned:
Initially, I was very sceptical about AI “agents”.

“Sure bro, an AI decided to launch a token and start shitposting about it on Twitter…”

I dismissed it as another short-term narrative built to pump tokens.

However, the tokens & tweets are a small part of AI agents.
Let's start with the tokens since that's why most of you are here.

AI Agent tokens typically fall into two categories:

· Governance tokens that let holders influence the project's direction (ai16z)

· "Attention" tokens - essentially ways to speculate on an agent's popularity
Love it or hate it, Twitter is where the attention is and tokens are how you financially capitalize on that attention.

But Tweets & tokens are just modules from a massive pool of possibilities that AI agents can tap into.

So with tokens covered, let's cover tech.
What separates AI agents from simple chatbots is “actions”.

Actions are how the AI performs tasks on external systems.

The most common action we’ve seen so far is posting on X. Image
The developer defines specific actions that the AI agent can call.

The AI agent can then trigger these predefined actions whenever it thinks it should do so.

It’s up to the LLM to decide what actions to call (if any) when it gets prompted.
Actions can be on-chain or off-chain.

For example, ai16z's Eliza framework has on-chain actions for:
· EVM: bridging, swapping, transferring, etc.
· Solana: pumpfun, swap, transfer, etc.

Off-chain actions:
· Generating videos
· Posting Tweets
· Replying in Discord
etc... Image
The agent decides to call actions based on the given input.

For example. A user might directly send a message to the agent saying:
· “Deploy a token”.
· “Create an ERC20”.

The AI/LLM interprets the message and decides if it should call any of its known actions. Image
Actions can be anything you can write in code.

Some examples: submitting blockchain transactions, or replying in Discord.

For example, I made a 2-minute demo video of an AI agent calling onchain actions:

I also built an AI agent that you can talk to in a Discord voice channel.

These showcase how agents can express themselves in different forms.

(both of these are fully open-source if you want to build your own btw).

The agent interprets the input and decides what action to take, if any.

There is also a system prompt the agent accepts that defines its personality.

For example, I made 10 agents here that act like celebrities:

What makes these agents autonomous is how their input is provided.

For example, rather than a simple chat interface, it might:
· Respond to Tweets/Messages
· Listen for blockchain events
· Periodically run with a given prompt
More advanced agents like @aixbt_agent provide the agent with large databases that help them make better-informed decisions.

Using vector stores, agents can quickly search through large datasets across thousands of files, e.g. tweets, market data, etc. Image
These databases also enable agents to track new information, including their previous actions.

For example, ai16z's Eliza framework comes with built-in memory management.

This allows you to create, search through, and delete memories. Image
TLDR: “Onchain AI agents” are:

· Onchain: control a wallet + can read & react to blockchain-related events.

· AI: An LLM decides what actions to call based on the input.

· Agents: inputs can come from various sources and the AI remembers it’s previous actions.

Onchain parts: Image
AI agents - bullish or BS?

IMO, bullish.

AI controlling a wallet unlocks a lot of fun and interesting use cases.

It might even be an early look at how we interact with the internet in the future.

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Jarrod Watts

Jarrod Watts Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @jarrodWattsDev

Nov 22
Ethereum block production is extremely centralized.

95% of Ethereum's blocks are made by 2 entities, leading to the potential for transaction censorship.

EIP-7805 introduces FOCIL, a way to preserve Ethereum’s censorship resistance.

Here’s how it works: 🧵 Image
Justin Drake's Beam Chain proposal includes 9 major upgrades.

Upgrade #1 is to improve censorship resistance.

But why? Isn't Ethereum meant to be decentralized?

Let's quickly cover how the block-building process has evolved to where it is today. Image
By staking 32 ETH to run an Ethereum validator, occasionally, you get to propose the next block in the chain and get rewarded.

In theory, this means block creation would be decentralized amongst the validators.

So, how do 95% of Ethereum blocks get made by 2 entities...? Image
Read 13 tweets
Nov 19
Beam Chain was the biggest announcement at Devcon, introducing 9 major upgrades for Ethereum.

But most people still don’t understand them...

So, here are 9 tweets to explain the 9 upgrades: 🧵 Image
Beam Chain is Justin Drake's proposal to redesign Ethereum's consensus layer (CL).

Before we dive into the specific upgrades, I'll briefly overview what the CL actually is.

Each Ethereum node runs two components:
1/ Execution Client
2/ Consensus Client Image
The consensus client is what nodes use to “agree” on what the next block in the blockchain is.

You can stake 32 ETH to add a third component, a “validator”, to earn rewards for proposing new blocks via the proof-of-stake algorithm. Image
Read 17 tweets
Oct 22
Abstract just announced Panoramic Governance.

It is a proposed way to reward both the users and app developers bringing the most value to the chain.

Here's how it works (in simple terms) 🧵 Image
Panoramic Governance (PG) consists of two parts:

1. Activity Driven Loop
2. Emissions Driven Loop

(1) Incentivizes users for their active participation.
(2) Provides rewards for the best app builders.

(1/12)
Before we dive into these two parts, we need to first understand how L2s work today.

A critical component of each L2 is the sequencer.

The sequencer is responsible for both:
- Processing all L2 transactions.
- Posting transaction batches to Ethereum L1.

(2/12) Image
Read 13 tweets
Aug 16
Account abstraction is finally gaining the traction it deserves over the past few months.

So, why not build it natively into the protocol?

That's exactly what @AbstractChain is doing... and I spent the past week learning about it.

Here's exactly how it works (a thread): Image
As a quick recap, account abstraction enables the use of smart contracts as wallets (as opposed to EOAs).

It provides a standardized way for smart contracts to act as a wallet and handle transactions.

EOA: public/private key pair.
Smart Contract: any valid EVM code.

(2/24)
The way it works on Ethereum is by having two separate mempools (waiting areas) for transactions.

These are:
1. Regular transactions submitted via EOAs.
2. UserOperations (a new type of pseudo-transaction that smart contract wallets can handle).

(3/24) Image
Read 24 tweets
May 8
Vitalik just proposed EIP-7702.

It's one of the most impactful changes Ethereum is going to have... EVER.

So, here's everything you need to know about how it works and how we got here:
The new EIP-7702 proposal is surprisingly short! Leaving some people confused as to how this is going to work.

To understand 7702, there are first three other proposals that it mentions that we'll walk through:
1/ EIP-4337
2/ EIP-3074
3/ EIP-5003

(1/23)
Let's start with the goal of *all* of these proposals.

EOAs ("normal" accounts) on Ethereum suck. They're risky and have very limited capabilities.

Account abstraction allows you to use a smart contract as your account to add more features & safety to address this.

(2/23) Image
Read 26 tweets
Jan 24
Ethereum L2s have now reached over $20B TVL.

But users still have almost no way of performing cross-chain L2 transactions.

I spent the past week reading about how different kinds of sequencers (and Polygon's AggLayer) are being built to solve this.

Here's everything I learned:
First, what even is a sequencer?

Rollups consist of several components, but all of them include some kind of sequencer that performs 2 critical tasks:

1/ Read transactions from the L2 mempool & execute them
2/ Batch transactions together, and send the batches to L1 Image
Today, most rollups:

1/ Use their own sequencer, responsible for reading transactions exclusively for their L2.

2/ Use their own L1 rollup contract, where the batches of transactions are stored exclusively from their L2. Image
Read 25 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us!

:(