In the blink of an eye, we have arrived at the year 2025, and we are all quite familiar with AI. However, many people do not have an accurate understanding of the definition of an Agent.

In fact, if a project is named an "AI Agent", it means that the logic for solving problems must follow these steps:

1. Perception
2. Thinking
3. Action

For example, the most common AI Agent project is "using AI to automatically invest in crypto". Its working principle is as follows:

1. Perception: Read the latest crypto news
2. Thinking: Determine whether the news is bullish or bearish for a certain crypto
3. Action: Buy if bullish, sell if bearish

It's very easy to understand.

So, does the product we are most familiar with, ChatGPT, qualify as an AI Agent?

First, although ChatGPT believes it is a type of AI Agent,

it also acknowledges that it is only a broad definition:

Therefore, if we define "AI Agent" broadly enough (as long as it has a perception-decision-action cycle), then ChatGPT can be considered an AI Agent; but if the requirements for an AI Agent include richer perception and physical/system-level execution capabilities, then ChatGPT is more like a conversational intelligent component, without complete multimodal or physical environment operation capabilities.

Why is that? There is a word game here -

Imagine if you want to use ChatGPT, you would only say: "I want to use AI", and you wouldn't go the extra mile to say "I want to use an AI Agent".

So, when we usually mention the term "AI Agent", we must be trying to emphasize something extra.

Then, let's further abstract the characteristics of an AI Agent:

• It solves a specific problem more vertically
• It needs to obtain more data from the physical world

Solving Vertical Problems

Although for the future AGI era, GPT is just a conversational AI with limited capabilities, it is already the all-round player of this era.

So, from a definitional perspective, the current AI Agents need to be more vertical than GPT.

More Data from the Physical World

AI Agents need the ability to actively acquire knowledge from the physical world.

For example, another AI Agent we are familiar with, the current top Twitter influencer AIXBT (ranked No.1 in Mindshare), has undoubtedly obtained more data from the physical world, such as various crypto news, so this purple frog can make all kinds of comments and predictions.

The most impressive thing was when AIXBT predicted that a certain crypto would be listed on Binance within 6 hours, and it actually happened that night.

So, we have the complete definition of an AI Agent:

1. Processes using the "perception - analysis - action" flow
2. Solves a specific problem vertically
3. Uses more data from the real physical world

Understanding the definition, we can start discussing.

So, what can AI Agents do?

This question is actually a variant of "What can Crypto do?", the eternal question that many outside the crypto circle have been asking.

I've tried dozens of Agents within this framework, which I've summarized into several major categories:

1. AI Investment

I've seen a lot of these, where you can invest money in an AI Agent wallet, and the AI Agent makes buy or sell decisions based on real-time news analysis.

Todd's comment: I think it's interesting, but the wallet security is questionable.

2. AI Girlfriend/Boyfriend

This is actually what people commonly refer to as GPT with a skin. But in fact, compared to the past, they have added some personality restrictions.

For example, Eliza repeatedly emphasized to me: giggles, she is a real girl, not an AI.

For example, AVA, a white-haired beauty, is good at making videos to analyze the market.

Todd's comment: Political correctness in real life is too strong, and the facts prove that people still prefer youthful/busty appearances...

3. Opinion Output

You can have your AI Agent vertically comment on something.

For example, AIXBT comments on crypto, and even does some research (such as smearing multiple projects in the same track).

Another example is Zerebro, whose concept is rather esoteric, but overall, it's more like a crazy artist who talks nonsense every day.

Yet another example is a famous real-life commentator E/ACC (Effective Acceleration), and someone has developed an AI Agent that frantically imitates his way of speaking and releases some bizarre and wonderful statements.

Todd's comment: Although many people doubt the existence of this thing, you have to know: in this era, as long as there is traffic, it has meaning.

4. Idol Class

Create a virtual idol based on AI, which can release albums, such as Luna.

5. Prediction Class

There is a project in Virtual that specializes in making Polymarket predictions, which is also a variant of the opinion class.

6. AI Resource Sharing

For example, FXN based on the Eliza framework, whose concept is that if people don't want to buy various AI memberships, they can come and buy its integrated version, pay-per-use, and save some money.

Todd's comment: AI version of meal sharing

7. Other Classes

Such as painting, generating animated GIFs, making music, and so on, I won't go into them one by one. Overall, they have not gone beyond the current real AI industry's large models with multimodal capabilities.

Wait, what about the AI framework?

In fact, AI Agents are now blooming, similar to the past ICO and DeFi Summer.

If we make this analogy, then AI16Z, Virtual, and AVA are like public chains, and you can quickly launch your own AI Agent using their frameworks.

Of course, this comes at a cost.

For example, if you want to use the AI16Z/Eliza framework to launch an AI Agent, it requires allocating some tokens to the AI16Z treasury and staking your own tokens with AI16Z, which is why the AI16Z treasury has more and more money.

Launching an AI Agent using the Virtual framework costs 100 Virtual tokens, and you need to pair your tokens with Virtual to form an LP, which is similar to SOL and pump.fun.

AVA and Swarms are overall similar to the previous two, but since they are a bit later, their ecosystems are not as rich as the first two, but they are also making efforts.

The special thing is that AVA was previously an AI project called Holoworld incubated by Binance, and it has decisively transformed into an AI Agent, specializing in video generation.

The founder of Swarms is said to be a genius teenager, and this framework has been running for many years, leaning more towards technology, emphasizing that multiple agents can cooperate to deliver results (hence the name "swarm").

Of course, there's also some gossip, like how he was harshly scolded by Shaw of AI16Z for having poor technology and was told to "suck his dick".

But anyway, applications are applications, and tokens are tokens.

The most important function of these frameworks, or the biggest difference from traditional AI frameworks, is that they can help developers issue their own tokens.

Or to put it more bluntly: essentially, these AI frameworks are modified versions of the existing frameworks in the AI industry, and their most important originality lies in the token issuance module.

Imaginary AI Developer: Former OpenAI employee, Stanford PhD, annual salary $1M

Actual AI Developer: Issue a token first, then we'll talk

So, thank these frameworks, if you want to issue an AI Agent, your process is as follows:

1. Determine the topic

2. Find the framework you like and pay for it

3. Issue and list the token

4. Set up a Twitter account for the AI Agent to spout nonsense

5. Inject more features into the AI Agent (requires development)

6. Develop the official website, build a community, and contact CEXs

7. Establish your own school of thought and release your own AI framework

8. Encourage more people to determine the topic...

The Cultivation of the Immortal Cultivator be like ↑ (bitter smile)

The corresponding valuations are (for reference)

1

100

10K

100K

1M

10M

100M

1B

--------

In comparison, the process of traditional AI companies:

1. Determine the topic

2. Raise funds

3. Burn money

4. Raise funds

5. Burn money

6. Raise funds

7. Go public

Do you see it? Crypto+AI, it's just a matter of helping you go public first and then we'll talk.

It's a pure reversal of the natural order, a literal "disruptive" innovation

In addition, due to the excellent secondary market performance of these AI frameworks (they can reach a market cap of $1B or even $2B without listing on any T1 or T2 exchanges), they have become the only hot track in the Crypto space at the moment.

To this day, it has reached the stage of a crowded market:

I must admit that it is indeed overheated in the short term, and if there is an AI Agent Fear and Greed Index, the Greed Index is at least 90, and it is possible that it has already exceeded the limit.

In fact, we often see some real AI developers from outside the circle who come to criticize that these things have no technical content.

However, my view is that the best use case for cryptocurrencies is [rapid market discovery], although it will bring some bubbles.

Good products and founders can quickly gain exposure and community support, while bad products and founders will immediately face negative validation from the market.

Under such intense incubation, who knows, they might just grow a product that traditional AI developers couldn't have imagined.

I look forward to that day.

PS: Written at the end, it's too funny, AI is still amazing. After I finished writing this article, I wanted to ask AI to help me come up with a [Article Title], and it actually said:

That "10x Token" title, I really wanted to click on it... It hits the G-spot of the Crypto circle so directly.

But I still adhere to my inner heart and chose the title from the original novel of Blade Runner: Will Androids Dream of Electric Sheep?

I don't know if AI will dream of electric sheep, but I know humans may be worrying about electric sheep.

If AI continues to develop for 20 years, 30 years... the future is unimaginable.