Original author: Haotian (X: @@tmel0211)
After waking up, many friends asked me to look at manus, which claims to be a truly universal AI Agent globally, capable of independent thinking and planning to execute complex tasks, delivering complete results. It sounds very cool, but aside from the anxious voices in many social circles worrying about unemployment, what will it bring to the explosive growth of web3 DeFi scenarios? Here are my thoughts:
1) About a month ago, OpenAI launched a similar product called Operator, where AI can independently complete tasks such as restaurant reservations, shopping, ticket booking, and food delivery in the browser. Users can visualize supervision and take control at any time.
The emergence of this Agent did not spark much discussion because it is driven by a single model and follows the framework of tool invocation. Once users realize that key decisions still require intervention, they lose the idea of relying on it to execute tasks.
2) On the surface, manus seems not much different, just with more application scenarios, including resume screening, stock research, real estate purchasing, etc. However, the difference lies in the underlying framework and execution system. Manus is driven by a multimodal large model and innovatively adopts a multi-signature system.
In short, AI needs to mimic human execution of the PDCA (Plan-Do-Check-Act) cycle, which will be collaboratively completed by multiple large models, each focusing on specific links. This not only reduces the decision-making risk of a single model executing tasks but also improves execution efficiency. The so-called "multi-signature system" is actually a decision verification mechanism for multi-model collaboration, ensuring the reliability of decisions and execution through the joint confirmation of multiple specialized models.
3) In comparison, the advantages of manus are clearly highlighted, and the series of operational experiences showcased in the video demo indeed provide an extraordinary experience. However, objectively speaking, the iterative innovation of Manus over Operator is just the beginning and does not reach a revolutionary significance.
The key point lies in the complexity of the tasks it executes and the fault tolerance and success rate of the large model's delivery results after non-unified standard user input prompts. Otherwise, following this innovation, can the web3 DeFi scenarios be immediately mature applications? Clearly, it cannot:
For example, in the DeFi scenario, if an Agent needs to execute trading decisions, there must be an Oracle layer Agent responsible for collecting and verifying on-chain data, integrating and analyzing data, and monitoring on-chain prices in real-time to capture trading opportunities. This process poses significant challenges for real-time analysis; a trading opportunity that was valid a second ago may no longer exist by the time the Oracle large model transmits it to the trading execution Agent (arbitrage window);
This actually exposes the biggest weakness of such multimodal large models in making execution decisions: how to connect to the network, invoke and analyze Real-Time level data from the chain, and then analyze trading opportunities to capture trades. The networked environment is relatively manageable; many e-commerce websites do not have real-time price fluctuations, which does not significantly disrupt the dynamic balance of multimodal collaboration. However, on-chain, such challenges exist almost constantly.
4) Therefore, overall, the emergence of manus will indeed stir up a wave of anxiety in the web2 field, as many repetitive clerical and information processing jobs may face the risk of being replaced by AI. But let them worry about their own issues.
We must objectively recognize the impact of this on the promotion of DeFi application scenarios in web3:
It must be acknowledged that the significance is indeed substantial, after all, the LLM OS it proposes and the concept of Less Structure more intelligence, especially the multi-signature system, will provide great inspiration for the integration of DeFi and AI in web3.
This actually corrects the major misconceptions of most DeFi projects; one should not immediately think of relying on a large model to achieve complex goals like autonomous thinking and decision-making for AI Agents, as this is fundamentally unrealistic in financial scenarios.
The realization of the true DeFi vision requires solving complex issues such as the capability limits of individual AI models, the atomic guarantee of multimodal interactive collaboration, unified resource scheduling and allocation of multimodal systems, and system fault tolerance and failure handling mechanisms.
For example: An Oracle layer Agent is responsible for collecting on-chain data and analysis, and monitoring prices to form effective data sources;
A decision layer Agent analyzes and assesses risks based on the data fed by the Oracle and formulates a set of decision-making and action plans;
An execution layer Agent executes based on the various plans provided by the decision layer, considering actual conditions, including gas fee optimization, cross-chain status, transaction sorting conflicts, etc.
Only when this series of Agents are all simultaneously powerful and a large system framework is established can a true DeFi revolution be initiated.
What is the real impact of manus on the web3 industry? It is not a technical crush, but a spiritual and soul blow! To be honest, I have held back some thoughts for a long time, and I can't help but express them:
1) I originally did not understand why people in the web2 industry looked down on web3 AI Agents, but when web3 AI Agents are in disarray, looking at the endless stream of technological and application innovations in the web2 field, I have to recognize a fact. Our vision of AI + Crypto is not wrong, but the current web3 field is filled with a lot of opportunists, with many garbage projects playing the long-term builder card while engaging in meme schemes to exploit others;
2) I initially wanted to say that we need to give the market more confidence, that innovation is always on the way, but over time I found that after a pile of projects emerged, inflating hundreds of billions of dollars in bubbles, when the bubble bursts, it will not only harm every long-term holder but also hurt some practitioners who may still harbor a glimmer of Build ideas amidst the chaos. Continuous dumping is destroying the foundational confidence in the industry. It is not just the zeroed coin prices that are being cleared; perhaps a large number of originally passionate entrepreneurial teams are also being wiped out; the situation of bad money driving out good has no winners!
3) We may have all overestimated the innovative capabilities of web3 AI Agents. In a short time, we have seen AI Agents' autonomy, independent trading decisions, AI Agent NPC embedding, AI Agent metaverse modeling interactions, and other AI + Crypto innovation dreams all attempted. The result, however, is that most of the plans and visions are utopian self-talk, and many proposals were not even assessed for feasibility before being put forward. The result is a grand and inflated vision, and after a quick run-through, the coin prices have collapsed, and the originally outlined technical vision and roadmap have also become irrelevant; honestly, forget about the big things coming, just being a good follower of web2 innovation is already quite good.
4) I was originally proud of the unparalleled attractiveness of web3's Tokenomics and its ability to attract excellent talent from web2, but seeing the teams behind DeepSeek, Yushu Technology, and today's trending manus, most of them are PhD graduates from Tsinghua, Peking University, the Chinese Academy of Sciences, etc., in computer science, electronic engineering, and other fields. Then looking at the developer teams in the Crypto AI circle, they are either web2 developers who have fallen back to web3 looking for shortcuts, or they are habitual scammers in the web3 field, or they are technically capable but not receiving the attention they deserve as marginal developers in web3. When it comes to "talent," how can we compare the innovation speed with web2? If pure Tokenomics and issuing tokens is the source of all Build motivation, it is also the root cause of killing everything;
5) I originally thought the biggest dilemma for web3 AI Agents was time, as if time could heal everything. But now I know I was wrong. If the underlying value creation mindset is not reshaped, time will only bring about the next bigger bubble and will not change anything. Now, web2 has brought about the iteration of multimodal over unimodal, introduced the innovation of separating decision-making and action, and brought about the leap of LLM OS + open-source combination as an operating system. The frameworks and standards of web3, DeFi, GameFi, Metaverse, chainization, and other innovative build propositions and path directions have clearly pointed the way. I look forward to seeing what kind of team can break through the encirclement first and bring a glimmer of life to web3 AI Agents.
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。
