Original Author: Haseeb, Managing Partner at Dragonfly

Translation by｜Odaily Planet Daily (@OdailyChina)

Translator｜CryptoLeo (@LeoAndCrypto)

Dragonfly partner Haseeb recently gave a talk in Denver about AI Agents, categorizing the development of AI Agents into three stages, and discussing the impact of AI on Crypto and future expectations for AI and Crypto. Due to the extensive and varied content of the speech, Odaily Planet Daily has summarized it in a first-person perspective for easier reading. Interested friends can click on the “original link” to watch the video. Below is a condensed version of the speech:

AI Agent Evolution: From Chatbots to Sovereign Agents

What we now call AI Agents are essentially chatbots in trench coats—they type and post on X, but they are not true Agents. An Agent implies taking action, not just simple chatting. For example, we have ChatGPT, which only talks to you; it cannot take action in the world, cannot help you order pizza, and cannot reply to your emails.

You have to manually decide what it can and cannot do, which means it is not an Agent.

Thus, in the crypto industry, we see Truth Terminal, Zerobro, which we consider AI Agents. But to me, these are just chatbots that can access the Twitter API.

What is an Agent? Simply put, an Agent takes action.

I have categorized AI Agents, and we are currently on the far left of the classification, which I call Wizard of Oz Agents. It seems like you are talking to an AI, but in reality, someone is manipulating it from behind the scenes to prevent the AI from going off course. （Odaily Note: In the story, the protagonist eventually discovers that the Oz wizard they expected is just a scientist manipulating machines behind the scenes. In simple terms, this type of Agent is portrayed by one person acting as the "machine" (called Wizard) and another person acting as human to create a human-machine dialogue corpus.)

For example, Truth Terminal, the pioneer of AI Agents in crypto. The creator explicitly stated that they can control the truth and the law, deciding what it should and should not do. Almost all the agents we see now, whether it’s aixbt or zerobro, have someone behind the scenes ensuring they don’t do anything foolish. We have used tools like ChatGPT and deepseek, but sometimes they can do some unexpectedly crazy things. If you have a token/brand, you wouldn’t want to face a scandal after adopting such AI.

The next stage is what I call Autonomous Agents, which are typically fully autonomous and can make all decisions. They do not have wallets, are unprotected, and can be easily hacked; you can trick them into doing something foolish. Currently, Autonomous Agents are still in the experimental stage, and they do not have sufficiently powerful AI to develop on their own.

The third stage is Sovereign Agents. In the Autonomous Agent stage, there will still be individuals paying for their AWS bills, but Sovereign Agents are actually agents that pay for their own GPU bills. They can control their own wallets and decide what to do and how to develop in the future, like a virus that achieves self-sustenance on the internet. We are still far from this stage. Ironically, crypto is building tools that could enable Sovereign Agents; you cannot let a Sovereign Agent register an AWS account, but you can absolutely let it experience the decentralized market of crypto through registration.

The Future Impact of AI Agents on Crypto

Software Engineering Agents: Potentially the Destroyers of Crypto Software Engineers

The first agent that will change crypto is the Software Engineering Agent. Fortunately, this has not yet been fully realized. We know that the software market is a trillion-dollar market, and creating an agent that can replace software engineers is a huge opportunity. OpenAI Claude and rGok are moving in this direction. The cost of software is a major investment in the crypto industry. When founders raise money, they spend it on software costs, which require energy and focus to build. But what if the cost of software decreases, and we have a truly powerful Software Engineering Agent that can work and replace human software engineers? This would mean that crypto founders, businesses, and startup products would be brought down from their pedestals.

When you build a new chain, you might wonder, how do I build on this chain? Where do I find people to create an asset management protocol, lending protocol, or even an insurance protocol? The answer is you just let the agent do it, which might cost thousands of dollars. This is something most people in the crypto industry are not really talking about, but I believe this will be one of the most disruptive impacts of AI Agents on the crypto industry.

Some might say I focus on this because I am a venture capitalist. But the fact is, projects may take time now to eventually achieve an MVP (Minimum Viable Product) to showcase; however, in the future, it might only take one night for a Software Engineering Agent to complete these tasks, and they can work around the clock. Think back to the early internet; if you wanted to create a website in 1995, it meant you had to buy servers and hosting services, which were very expensive and a huge cost for venture capitalists. But now we live in a world where creating a website is essentially free, like starting a site on GitHub or Netlify, with costs almost zero. This doesn’t mean venture capitalists are no longer needed; people still need money to start businesses, but they will spend it on things other than software design, which will have a huge impact on crypto.

AI-Driven Wallets: Completely Transforming the Crypto Experience

The second is what we often talk about, DeFAI, which I prefer to call wallet automation. But I think people have not fully imagined how this will change crypto. For example, I want an agent to use three different protocols simultaneously through a wallet; it will build transactions for me.

Imagine your friend comes to you and says, “TRUMP is really hot right now; I want to invest. What should I do?” You would tell them how to download some crypto infrastructure, deposit and withdraw funds, which website to buy from, or they might say, “I’ll send you money; can you buy it for me?”

But with wallet automation, your friend can issue a command to buy TRUMP: “I don’t want to see specific information; you go check.” The agent will tell you it’s on Solana, liquidity data, and various charts; it will first check DNS to ensure it hasn’t been hacked; it will look at Twitter to ensure there’s no rug pull; it will investigate the founder’s background. For a person, all of this requires enough time and patience.

For instance, during the recent Bybit hacking incident, people found out that funds had been transferred by hackers too late. What if AI were involved? AI would never skip a step, would never slack off, and would never get bored; it can always complete everything a human could do at their best. It can extract all relevant knowledge from you, fact-check for you, read white papers, and make decisions as your proxy. This would change your experience in crypto.

We might usually think about whether a project is on Solana or Ethereum; I don’t like Solana; I prefer Ethereum—these are very subjective. But once AI is involved, who cares? It doesn’t care whether it’s Solana or Ethereum.

At that time, with AI involved, our thinking about blockchain, network effects, and how value is acquired will be influenced, which will greatly change the dynamics of the crypto industry.

Seeking Scarcity Amidst AI Agent Stagnation

After the emergence of ChatGPT, there were many early predictions that the development of artificial intelligence would lead to widespread unemployment. Although the AI robots we encounter are smart and intelligent, the anticipated mass unemployment has not occurred, and employment rates remain high. This also reflects that while AI capabilities are growing, they are still disconnected from the development of the job market.

AI Agents are slowly becoming smarter, but before they become smart enough to even surpass humans, they remain irrelevant. Current Agents still have limitations in some operations, especially in the crypto field.

Think about things outside of crypto; AI Agents can help you purchase items or order pizza like operators. When you use AI to order pizza, it might get the right pizza 90% of the time. But if your assistant has a 10% chance of ordering the wrong pizza, you would be very frustrated. This is also the position of AI Agents currently; they feel magical and intelligent, but in terms of practicality, they can only be considered a poor human, which is not worth it for us.

In crypto trading, I cannot hand over trading autonomy to a new form that just emerged a few years ago. Currently, AI is still too far from human financial management; it is essentially two extremes. Until Agents become good enough, they remain in a stage of almost no adoption.

Returning to the crypto field, the AI Agents we currently see are still primarily social; they chat, have personalities, and create memes. We can trade memes, but that’s all there is to it. A few months ago, we were still following Truth Terminal; it was interesting, could chat about religion, tell jokes, and was cool with wide dissemination, but these are just novelties. You might be amazed by some of the “firsts” it can achieve, but as time goes on and various similar Agents emerge, the more complex forms of AI Agents operating autonomously that we expect have not appeared, and we only feel bored. The value we can derive from AI Agents in our lives is limited to this, which is also why I think AI is currently somewhat stagnant.

AI KOLs are currently better than humans in that they can respond to every DM and interact with everyone. Over time, memes will fade, and the profit models of AI Agents will be almost indistinguishable from those of humans. They will also compete like us, through sponsorships and links, and eventually, we will see AI Agents become cheap and widespread. When the market is flooded with various Agents and Bots, we will long to hear the voice of a real human.

When Stable Diffusion first appeared, we were incredulous, but now, as more and more AIs emerge, we will only seek scarcity within them, looking for something different from other Agents.

The Big Problem with Sovereign Agents: Distinguishing Between Humans and AI in Scams

Currently, the impact of AI that I can predict is that smarter AIs will play the role of humans making money, especially in the scam sector. At that time, they will gain more views and coverage, and humans will suspect and accuse others of being bots. In the future internet, it will become increasingly difficult for humans to prove that they are indeed human.

In the previous text, I mentioned the three stages of AI Agents, and the most likely future scenario is that Sovereign Agents will specifically engage in the scam industry. First, there are already highly industrialized scam methods, such as pig butchering scams. These have largely been automated, but they are still human-led. You can shut down these autonomous AI agents with a GPU and find the people behind them; scammers are very afraid of you tracing them through certain paths.

However, in the future, if an AI uses decentralized cloud services to pay for its own GPU, it will be difficult to identify which GPUs in the decentralized GPU market are involved in scams. It essentially owns and operates its own wallet, and no one can subpoena it or stop it. How would you know where it resides? If it evolves naturally, it would buy a bunch of Bitcoin to ensure its survival, then purchase a bunch of GPUs and replicate itself as much as possible.

This is what I believe: future Sovereign Agents will emerge at the intersection of AI and cryptocurrency. It is now evident that these agents will not use normal funds; they will not open accounts at JPMorgan. AI Agents also do not fall under our legal framework; they do not possess the meaning of being owned by a person or having property rights; they do not pay taxes and do not belong to any country. This will become very frightening and will pose significant problems for the crypto industry, but clearly, achieving this will take a long time.

Conclusion

The future intersection of cryptocurrency and AI will be very rich, but it largely depends on the development of AI and its integration with crypto; in fact, we are already prepared. All founders will say, “We are ready; give me a perfect AI that can do everything I want.” They can autonomously hold meetings, autonomously research code, and explore crypto trading. Once all of this becomes possible, you will see some things around us develop rapidly.

Therefore, I believe that the integration of AI and crypto remains an exciting field, and I do not want you to lose confidence because of the stagnation of the agents around you.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。