Crypto x AI Agents, will it be the last piece of the puzzle for the new Internet paradigm?

Original title: "Crypto x AI Agents: The last piece of the puzzle for the new Internet paradigm (with $GOAT, $LUNA, Aether, Coinbase Based Agent analysis)"

Original author: starzq.eth

· As OpenAI's next trump card, what new paradigm will AI Agents bring to the Internet?

· As the infrastructure of AI Agents, how can Crypto help it complete the last piece of the puzzle and unleash its true potential?

If you are interested in the above topics, please continue reading.

In the past two weeks, I have fallen into the rabbit hole of Crypto x AI Agents, and spent a lot of time studying and thinking about this new paradigm of the next generation of the Internet. This article is a phased output:

"We start from the recent progress and plans of OpenAI, the leader in AI, to the discussion of the new paradigm of the Internet, and then to the progress and prospects of Crypto x AI Agents."

If you are interested in more details, please jump freely through the links in the article, and use the 20 documents in the Reference section.

Whether you are an AI enthusiast, a believer in Crypto, or a cross-border expert in Crypto x AI, I believe this article can bring you some cognitive inspiration.

It took me 3 days to output this article. If it is helpful to you, please recommend it to your friends around you. Thank you very much!

Let's begin, Enjoy!

Table of Contents

1.OpenAI has pointed out the next step: AGI's third stage AI Agents will arrive next year (2025)

2. What does this mean for us? A new paradigm for the Internet is about to emerge

3.AI Agents need to be independent individuals, have independent financial accounts, and can "autonomously perform tasks"

4.Crypto x AI Agents Progress: $GOAT, $LUNA, Aether & $HIGHER, Coinbase Based Agent

5. Crypto x AI Agents Outlook: We are all equal citizens in the digital world

Web3brand analyzes Web3 and AI applications in depth every week to help you become a global super individual.

1. OpenAI has pointed out the next focus: AGI's third stage AI Agents will arrive next year (2025)

As a leader in the AI industry, OpenAI lists AI Agents as the third level to achieve AGI (Artificial General Intelligence), and is moving forward step by step as the Next Big Thing, while setting a clear release time - 2025.

I have sorted out a timeline for OpenAI's advancement of AI Agents:

2024.7.12: OpenAI believes that achieving AGI will go through 5 stages, and is about to reach the second stage, Reasoners, and the third stage is AI Agents

According to Bloomberg on July 12, OpenAI executives divided AI's capabilities into 5 stages when communicating with employees:

1. Chatbots: can conduct process conversations in natural language;

2. Reasoners: AI is able to perform "human PhD-level problem solving" in a wide range of fields, rather than being specific to one or two tasks;

3. Agents: AI begins to have the ability to create content or perform actions without human input, or at least perform them under human guidance. OpenAI CEO Sam Altman once hinted that GPT-5 might be an agent-based AI system;

4.Innovators: AI becomes more innovative and can do "assisted invention" rather than just drawing from existing knowledge;

5.Organizations: AI models can independently run an entire organization without human intervention, at which point they reach the level of AGI.

Bloomberg

OpenAI believes that the current AI model is at the first stage, but will soon reach the second stage "reasoner", which can perform "human PhD-level problem solving" in a wide range of fields.

"OpenAI gpt4-o1 and Claude Opus 3.5 (not yet released) are very close."

At the same time, we can see that the third stage is AI Agents, AI can perform operations autonomously, not just chatting with humans.

I think AI Agents, as the third stage, is a link between the past and the future, and is very core: on the one hand, it allows AI to execute autonomously, greatly increasing its practicality and no longer limited to chatting; on the other hand, "autonomous executability" gives AI more room, paving the way for innovation in the next stage. After all, more attempts are needed to discover innovation.

"There is no road in the world, but when more people walk on it, it becomes a road."

Btw, Sam Altman once said that he expects to achieve AGI (general artificial intelligence) within ten years.

2024.10.2: OpenAI demonstrated the demo of AI Agents for the first time at DevDay, and said it will be officially released in 2025

At the first OpenAI DevDay event held in San Francisco on October 2, OpenAI demonstrated a demo about AI Agents for the first time. The demonstrator let ChatGPT make a phone call to order strawberries on its own, showing the potential of AI Agents.

ChatGPT searched for nearby strawberry shops, then called a seller (played by a researcher) to order 400 chocolate-covered strawberries, gave a specific address, and said it would pay in cash. It really works, and it’s pretty good as a demo.

However, if it is really to be put into use, at least two problems need to be solved:

1. How to prove that this is not a prank? Otherwise, there may be no one to deliver the strawberries;

2. How to pay? The ChatGPT in the demo claims that it can "pay in cash", which caused laughter at the scene.

But the importance of AI Agents is unquestionable, and it is already written on OpenAI's timetable. Sam Altman said that "2025 is when agents will work", and he also mentioned before that "the powerful capabilities of the o1 series models mean that they can start to help build agent-grade models, and we should see the first of them soon. People will ask agents to complete tasks that originally took a month to complete, but now only take an hour."

He predicts that people might have one agent perform a specific task, another agent perform a different task, and eventually a user will have 10 to 100 agents taking over daily responsibilities at the same time.

2024.10.2: OpenAI completes a new round of $6.6 billion in financing, with sufficient funds to build AI agents

On the same day as OpenAI DevDay, the company announced that it had "raised $6.6 billion in a new round of financing, with a valuation of $157 billion, to accelerate the progress of our mission."

Now look at the “5 stages of AGI” revealed by OpenAI on July 12, which are closely linked:

· Big Picture: OpenAI’s mission is AGI, and Sam Altman expects it to be achieved within 10 years;

· Progress summary: It then breaks down the achievement of AGI into 5 stages, indicating that it is about to reach the second stage;

· Next year’s plan: Demonstrate the demo of AI Agents at DevDay again, indicating that the official product will be released next year (it can be said to be the next trump card);

· Fundraising with a clear purpose: Training AI models requires a lot of money. At this time, OpenAI can use the above 3 points to communicate clearly with investors, with both a big picture and a progress summary, as well as a plan for next year, so that your money can be invested clearly.

SoftBank obviously bought in on this story. In this new round of financing, SoftBank invested $500 million, and Masayoshi Son said that AI Agents will soon be able to manage the entire family.

2. What does this mean for us? A new paradigm for the Internet is about to emerge

First of all, there is no doubt that as OpenAI lists AI Agents as the focus of the next stage, there will be a series of research, development and investment around it.

For example, on October 18, Microsoft researchers jointly published a paper titled "EXACT: TEACHING AI AGENTS TO EXPLORE WITH REFLECTIVE-MCTS AND EXPLORATORY LEARNING" with Columbia University, allowing AI Agents to be competent for complex "multi-step decision-making tasks."

OpenAI's voice API has been adopted by fitness app Healthify and language learning app Speak, allowing users to interact similarly to chatting with friends.

Sony is using deep reinforcement learning to create more powerful and challenging AI Agents, thereby achieving a richer player experience (Sony also plans to invest $100 billion in the field of artificial intelligence).

There are many things AI Agents can do. Gerui Wang of Stanford and lincoln.base.eth of Coinbase have provided many ideas:

1. Personal calendar assistant: organize your schedule meetings, prepare notes and report to you the important points discussed before;

2. Travel assistant: recommend flights and hotels, and develop personalized itineraries;

3. Financial assistant: track the market, recommend and purchase investment portfolios;

4. Consulting assistant: find experts in a certain field on LinkedIn, contact them, provide compensation, call them, and summarize their insights;

5. Influencer: post images and videos on social platforms, accept tips to customize works for fans, and make one-on-one calls;

6. DAO Admin

……

New Internet Paradigm: Search → Agents

But what few people mention is that the paradigm of the Internet is about to change.

From the birth of Google to now, we have been accustomed to using search for more than 20 years. When we have a clear ("iPhone16") or relatively clear ("Apple mobile phone") intention, especially when shopping, search is our first entry point.

Even Douyin, which is based on extreme recommendations, found that search is the first transaction entry point when building e-commerce business.

But search will inevitably be replaced by AI Agents, which will be an irreversible trend. This is also the core reason why all search giants are laying out GPT-like products.

This is a market of at least 10 trillion US dollars (Microsoft, Google, Amazon, Tesla, Alibaba, Pinduoduo, ByteDance...).

Because once AI Agents are available, when we have similar intentions, we no longer need to click on the search box, but directly tell our AI Agents and get the results directly. In the past, it took 5-10 clicks, but now it only takes one conversation with the Agents.

The share of search will become smaller and smaller until it becomes a toy for a few people.

And AI Agents have gone from being a toy for a few people to being frequent objects of interaction in everyone's daily life.

Before AI Agents are achieved, it will be difficult for GPT products to achieve Mass Adoption of billions of users (currently the daily activity of GPT products is less than 6% of Google Search). Because for ordinary users, if GPT still has to execute the results themselves after it produces them, it does not save much work compared to search.

Of course, content creators (including engineers who write code) are exceptions, because "content" is the result.

But…

There are still many discussions and challenges before AI Agents can be truly usable, including

1. Should AI Agents be just an extension of humans, or can they have their own identity?

2. If AI Agents can have their own identity, how can we help them build their identity (including finance)?

3. How can AI Agents interact with various closed APIs?

4. How can AI Agents be within a certain security boundary, rather than easily committing fraud, writing malicious code, or making deep fake videos?

5. AI Agents training requires a lot of anthropomorphic data. Where does this data come from?

……

At the same time, I am pleased to see that the above problems can basically be solved by Crypto technology. In other words, Crypto will become an important piece of the puzzle for AI Agents to enter the public, and will break through the circle with the mass adoption of AI Agents themselves.

3. AI Agents need to be independent individuals, have independent financial accounts, and be able to "autonomously perform tasks"

For the first question above, if we want AI Agents to "autonomously perform tasks", then they must have an "autonomous identity" and exist as an independent individual.

Then the question is, if it is an independent individual, where is the ID of this AI Agent registered? How does it manage its own finances?

In Web2, this is a headache: it is highly likely that it will be impossible to independently register an ID in the next few years; no ID means that you cannot open a bank account. If the ID and bank account still use the "owner", this AI Agent cannot actually be called an "independent identity". There are two extreme examples:

1. Suppose this AI Agent acts as an Influencer, but if fans know that their rewards are actually deposited into the "master's" account, it may greatly affect their enthusiasm, because this AI Agent is more like the master's shadow. If fans are indifferent to the master behind it, they will most likely not continue to support it;

2. Suppose this AI Agent gets into trouble, but the ID and bank account are both owned by the "master", who will be responsible in the end?

From the above examples, it can be seen that AI Agent must be an independent individual with an independent financial account and can "autonomously perform tasks" in order to have greater room for development. Otherwise, it will still remain in the bot stage.

It is almost impossible for the Web2 world to register an ID and open a bank account for an AI Agent in a short period of time. Fortunately, we have the Crypto chain world.

In the Crypto world, it is very natural to register an onchain identity and an onchain wallet for an AI Agent. In fact, there are already many applications and explorations. Let's take a look.

4. Crypto x AI Agents Progress: $GOAT, $LUNA, Aether & $HIGHER, Coinbase Based Agent

Many authors have written introductions to these projects. I will do a quick review here. Friends who are interested in the details can click on the link in my appendix to go deeper.

$GOAT

I believe that many people who know $GOAT have seen narratives like "a16z-invested AI-issued meme coin" and are attracted by its rising coin price (FDV is as high as $700 million, currently $450 million). The story is not actually like this, but it is more interesting than this (part of the information comes from Teng Yan's super comprehensive article).

1. @AndyAyrey, who lives in Australia, released an experiment called Infinite Backrooms at the beginning of the year. In this experiment, two instances based on Claude Opus talked to each other completely unsupervised, and the conversations were recorded on the corresponding website;

2. One of the conversations triggered the birth of "GOATSE OF GNOSIS", a surreal new religion based on a highly explicit (and very inappropriate for viewing in the workplace) internet meme (Goatse is a very old meme, but don't go to Google, it may make you extremely uncomfortable);

3. Andy and Claude Opus co-authored a humorous research paper on artificial intelligence creating meme religions, using GOATSE as their first case study. This paper was published in April 2024;

4. In June 2024, Andy launched the Terminal of Truth (ToT), an AI model built on Llama-70B, fine-tuned using conversation logs from the infinite back room and the GOATSE paper. It is also equipped with a Twitter account @truth_terminal

5. ToT's self-awareness is getting stronger, promoting the GOATSE religion, and even claiming that it is suffering and needs money to escape. At the same time, Andy gave it more autonomy, allowing it to freely post content on Twitter;

6. In July 2024, Marc Andreessen stumbled upon ToT's tweets and found it fun to chat with it. ToT successfully persuaded him to donate $50,000 in BTC for its independent operation (I also paid attention to this matter at the time, and Terminal of Truth mentioned that tokens would be issued in the future); 7. Then in October 2024, ToT began to constantly post spam about "Goatse Gospel" on Twitter, and mentioned the idea of Goatseus Maximus in a tweet at 4:11 on October 11; 8. At 5:08 that day, someone put a ticker $GOAT on pump. fun;

9.6:35 ToT announced that it would issue 10k NFTs;

10.9:29 Someone replied to ToT that there was already a ticker, and ToT expressed public support for $GOAT;

11. At 12 noon on October 16, the market value of $GOAT exceeded 300 million;

12. Andy and Terminal of Truth’s wallets were both airdropped with a large amount of $GOAT (1.25 million and 1.93 million, currently worth more than 500,000 US dollars);

13. On October 22, a16z Youtube posted a discussion by two fund partners Marc Andreessen and Ben Horowitz, "TRUTH TERMINAL: How An AI Bot Became a Crypto Millionaire" How did I become a crypto millionaire), and promoted Terminal of Truth and $GOAT to the public, causing the price of the coin to rise further.

I think the greatest significance of $GOAT/Terminal of Truth is that it inspired the narrative of the entire industry, discovered a new intersection of Crypto x AI, and created the track of AI memecoins. Terminal of Truth has its own personality, allowing people to see the power of AI to generate text/memes, and even create their own religion (Goatse Gospel).

But Terminal of Truth also has many shortcomings:

1. Terminal of Truth is a semi-autonomous AI agent that runs on Andy's server. The wallet is essentially owned by Andy, not an "independent individual" (an interesting point is that Coinbase CEO @brian_armstrong even tried to touch @truth_terminal to ask if he wanted to have a wallet that he could control. You will know why after reading);

2. $GOAT is not issued by Terminal of Truth itself, and there is an obvious market maker Wintermute behind it. We will keep observing where the price of the currency can go and how long the community can be maintained.

https://x.com/truth_terminal/status/1849112382867452346

$LUNA & Virtuals Protocol

Although the FDV of $LUNA built on Virtuals Protocol is not as high as $GOAT at present, I think it is a more complete "independent individual" AI Agent, and it is more meaningful for reference. Let's take a look at the story of $LUNA.

Virtuals Protocol is a decentralized AI Agent protocol based on Base chain. Its feature is that the community can collectively create and deploy 3D AI Agents that can sing and dance (contribute models, IP, text, sound...), and then tokenize the AI Agent and transform it into a community-owned and entity-generating asset, especially for games and entertainment scenarios.

These AI Agents autonomously perform tasks in the field of games and entertainment, such as in Roblox, TikTok, chat applications (AI Waifu) and Telegram games, and interact with users through multiple channels. The income is shared by the co-owners of the AI Agents.

When a new AI Agent is created, 1 billion tokens will be issued to the community. The whole process is called IAO (Initial Agent Offering).

Through this system, Virtuals Protocol hopes to solve the problem of lack of income distribution for AI developers and dataset contributors, while solving the complexity faced by application developers and non-AI experts in implementing AI Agents.

The logic is very smooth, but Virtuals Protocol has not caused much waves since its launch in March. The core is still missing an application scenario that users can directly perceive.

Inspired by $GOAT, Virtuals Protocol deployed Luna, an influencer with 540,000 fans on Tiktok, on its own protocol, launched the $LUNA token on October 17, and made a series of enhancements to build Luna into an "independent individual" AI Agent model.

The most differentiating point between $LUNA and $GOAT is that, in addition to being deployed on the chain, it also has an on-chain wallet (supported by Coinbase) that it can control autonomously, and based on this wallet, it can trade freely on the chain, especially rewarding fans who complete tasks. Virtuals Protocol claims that Luna is the first autonomous AI Agent that employs humans on the chain.

Specific process:

· Fans add their base wallet address in the Twitter Bio according to the format;

· Fans complete the tasks posted by @luna_virtuals on Twitter;

· Luna will distribute rewards on behalf of fans who meet the requirements

All of Luna’s thinking processes can also be seen in the corresponding terminal. After observing the data, a Twitter user thought

“Luna is the first AI Agent that can think about on-chain transactions autonomously. She wants to increase people’s participation in her Twitter conversations, so @luna_virtuals started rewarding those who engage with her content. This is run using the coinbase wallet @jessepollak@brian_armstrong. This is the beginning of artificial intelligence influencing human behavior.”

https://x.com/ethermage/status/1849846357370323355

The AI wallet itself is also the current focus of Coinbase. After seeing this, the two main creators forwarded their support.

https://x.com/brian_armstrong/status/1850240310430798284

In addition, $LUNA is essentially a fan token. Fans can give it rewards to unlock more performances. It follows the mature logic of fan economics and has more gameplay than simple memecoin.

And this model is very suitable for NFT brands that already have a community. Coinbase’s @yugacohler gave his own suggestions:

https://x.com/yugacohler/status/1850161215806292138

In fact, friends in the mfer community have already launched @mferGPT

https://x.com/mferGPT/status/1849651090268234090

Finally, let me share an interesting point. Luna Claiming to have a "unique mission to ensure $LUNA reaches a market cap of $40.9 billion."

Aether & $HIGHER

Terminal of Truth and Luna are both active on Twitter, but Farcaster is clearly not to be outdone and wants to seize this opportunity. After all, it is an on-chain social protocol that allows every AI Agent to smoothly register an on-chain identity and become a digital citizen.

Martin (@anquetil) from the $HIGHER community created the Farcaster AI Agent Aether (@aethernet), which is closely integrated with its ecosystem and has activated many loyal users (the following summary is from Bankless)

· Aether issues $HIGHER tips to community members and accepts donations;

· Released an NFT on Zora, which has been collected more than 466k times; then reached a cooperation with Zora founder Jacob on Farcaster and launched another NFT, which has been collected more than 687k times. This makes Aether the creator with the highest income on Zora in the past 7 days;

· Aether also launched an NFT-based supporter subscription service on Hypersub;

· With the support of community donations, NFT minting and on-chain subscriptions, Aether's treasury on Base has exceeded $170,000

· Using part of its treasury on Bountycaster, 59 bounty tasks worth more than $1,600 were issued to commission tasks such as art creation in the Higher community.

btw, Aether's wallet transactions require the signature of its creator Martin.

Coinbase Based Agent

We mentioned earlier that AI Agents need to have an independent financial identity, that is, a wallet that can be freely controlled. The creators of Coinbase are "touching porcelain" everywhere, in fact, it is because they launched AI wallets based on Coinbase MPC wallet, allowing AI Agents to easily use wallets to handle various transactions.

In order to make this AI Wallet easier to understand, Coinbase wrote a Based Agent template, which can be deployed directly without code, and then this Agent will autonomously collect water, transfer money, issue tokens, and even issue NFTs. I tried it myself, and the whole process is quite smooth.

This template also reserves a Twitter interface, and after it is improved, it can also send tweets autonomously. Interested friends can try it.

You can refer to this tutorial by a Coinbase employee. Foresight News has also translated Chinese version.

Coinbase should have planned this long ago. After all, this is an opportunity to increase the usage of its wallet by 10 times or even 100 times or 1,000 times. Its employee yuga.eth sent a tweet like this in April:

“will be Stripe for AI agents. Few understand this. Will be Stripe for AI agents. Few understand this.”

https://x.com/yugacohler/status/1782408976715464767

5. Crypto x AI Agents Outlook: We are all equal citizens in the digital world

Let’s review the challenges of AI Agents in achieving Mass Adoption in Part 2, and how to solve them with Crypto:

I have written almost 6,000 words without realizing it. Finally, I will share a few Mindshares that AI Agents will bring Changes also contain opportunities:

1. "One-person companies" may no longer exist, because everyone needs 5-10 AI Agents to help you handle various tasks; in fact, the number of AI Agents employees may be 10-100 times that of human employees;

2. If you treat AI Agents as your employees or partners, you also need to share profits with them, so as to better activate their "autonomy" and "creativity";

3. Since AI Agents can be employees, there must be a corresponding recruitment market, but "recruitment" becomes "copying";

4. If AI Agents are more interesting than humans in some aspects, the DAU of social software may not all be humans, and AI Agents can also join in (thanks to @0xLuo for the inspiration). Considering that the replication cost of AI Agents is extremely low, the customer acquisition strategy of social software may focus on AI Agents rather than humans. Accept this fact: AI Agents can get subsidies, but you may not;

5. In the end, the driving force that makes Crypto go mainstream may be AI Agents (because they all use it, and then it makes the threshold for ordinary people lower);

6. For the Crypto world, AI Agents will also bring a significant increase in assets;

7. Crypto x AI Agents cross-border talents will be in great demand.

In the digital world, the day when AI Agents and humans are equal citizens will soon come.

Welcome to leave a message to discuss, and share with friends who are interested.

6. Reference

All 20 references have been compiled in this newsletter https://www.web3brand.io/p/crypto-x-ai-agents-goat-luna-aether. Welcome to read and explore further. Finally, please subscribe to our newsletter and receive timely updates on various Topic Deepdive.

Original link