When AI Agent took over the chain of authority: How far is the new financial paradigm of Web 4.0

From ChatGPT to today 's various AI Agents have taken over complex tasks, and change is too rapid。

AI Agent is no longer just a chat robot. As they begin to make their own decisions, mobilize funds and even sign a "contract" with other Agent, a new financial world is taking shape in the chain. The trend of chain finance + AIAgent is defined by most as Web4.0。

On 21 April, the BlackBeats x-inspection Beating X-Cobox MuleRun co-sponsored an event in Hong Kong on a dialogue on financial trends around Web4.0, sponsored by Cobo, MuleRun and NOXCAT and attended by luxury cars Lotus, Web3 Application NOXCAT. The event focused on the deep integration of the encryption world with AI Agent。

The trend is coming

The event began, and the Block Beats editor-in-chief shared a set of data。

In the traditional sense, chain trading seems to be just a "little circle" of encrypted currency. However, crude oil holdings reached US$ 1 billion during specific geopolitical events on Trade XYZ, a chain-based commodity trading platform。

"Although $1 billion is a very small amount of data in the bulk of the world, it is no longer the same volume as other encrypted trading platforms. " Traditional finance’s perception of the chain has long been different from that of the “good-issue, good-dealing” shallows of the year, which, in his view, may be more “horrifying” than PayPal’s support for a currency deal。

Zhang Yong Lok, Senior Vice President, took the subject one step further。

He himself was a student at the EMBA University of Hong Kong, when he first got into school, using AI to ask questions and write about things; later, AI became a key member in case analysis to assist in decision-making; and now, everyone is trying to adapt business models with AI Agent。

When these things are on the chain, they change more thoroughly. Agent can make a deal on his own and even run a complex set of strategies for you。

So we're going to face a more fundamental question: how can we rethink production relations and business models when AI Agent can make decisions and act on his own

What is known is to turn this new trend into a subject that the public can understand and participate in. From Web3 to AI to Web4.0, more and more people are taking part in the discussion. These technical- and financial-oriented elements are also moving from minority issues to broader public discourse。

The observer, Sleepy, made it clear in a few words what Beating was going to do。

The cut is simple: technology and people。

From Stable Currency to Crypto, from AI to Brain Interface, Mars Industries... How do new technologies, trends and ideas affect this generation of young people? How does it affect this age, this society? It's a different medium that wants to make these things that seem so distant feel like things。

"I hope you know how it affects you and me and how it affects real life."

Keynote link

The CTO beam bright from MuleRun and the AI Growth Officer from Cobo, Brad Bao, spoke separately about their respective views on the AI Agent capacity and trust relationship。

A Web 4.0 infrastructure in the eyes of a safety engineer

The CTO beam from MuleRun highlights that to understand the future, it is first to understand the capacity composition of AI Agent。

He puts a mature Agent in abstraction to six dimensions:

- Mouth: Agent has been able to talk to you through mainstream IM software such as Twitter, Telegram, Discord, etc., and to respond to instructions。

- EYES/EARS: DATA SOURCES DETERMINE EVERYTHING. IT IS ACTIVE IN ONLINE SEARCHES, BUT MORE CRUCIAL IS THE ABILITY TO ACCESS PROFESSIONAL APIS, TO OBTAIN REAL-TIME, ACCURATE FINANCIAL INFORMATION, AND TO MOVE AWAY FROM DATA DELAYS AND MISSING FREE APIS。

- Brain: the ability to handle complex issues. It is able to handle complex tasks ranging from "price checking" to "precipitation of volatility, strategic retrospection" and to optimize token consumption and reduce costs in engineering。

Hand: Agent has a 7x24-hour stable cloud operating environment that allows for long-term monitoring missions and even public Internet IPs, which build and host a publicly accessible website or monitoring system。

-Memory: The better Agent has a long memory, the longer it takes, the better it gets to know you, the less repetitive communication。

- Knowledge networks: a shared knowledge network, which simply lends you the experience of others。

He showed an autonomous investment system built by a friend using MuleRun: captures news, financial, emotional data, and can automatically signal a silo。

At the same time, it also offers its own ideas for the future of AI Agent:

- The era of complete autonomy: the future of AI Agent can complete a full-cycle strategy without any intervention and competition will intensify between different Agent rivalries。

- Product differences are reduced and use becomes a source of excess earnings. The bottom models of each Agent converge with API capabilities, and what really opens the gap is how users define tasks, design workflows, utilize memory and knowledge sharing。

- Infrastructure shift to AI-native design. The future API is no longer an UI optimization for humans, but a call design for Agent。

How to build a smart economic trust layer?

When Agent started to use the money, who should be given the chain

Brad Bao, the AI Growth Manager from Cobo, discussed the topic。

“Not to give or not to give, but to give a discretionary, retroactive and revocable competence”

Web3 addresses asset ownership, while Web 4.0 addresses the economic relationship between Agent: what can be done, what can not be done, who is responsible for what is wrong。

The solution for Cobo is to introduce the concept of "Pact". A Pact contains:

User Intent: What to do

Execution Plan: Which chain and which address

Regulations: What conditions must be met

Termination conditions: How is it done? When

The contract defines an enforceable, auditable and revocable "trust layer" between humans and Agent. Together with MPC technology, Agent can run freely, but humans always retain ultimate control。

Their product, Cobo Agenic Wallet, has been accessed online in an earlier version, with core competencies including but not limited to:

- User creates a wallet with Pact binding for Agent

- USE OF MULTIPLE COMPUTING (MPC) ENCRYPTION TO ELIMINATE SINGLE POINT RISK

- Real-time monitoring of Agent behavior, abnormally automatic freezing

- Supporting humans to withdraw all clearances

@AgenticWallet+Pact = Turning "authorization" into "how to implement a contract together"。

Web 4.0 is not a machine that replaces human beings, but rather a trusted economic relationship between humans and AI Agent。

How long before the new financial form of Agent's influence comes

In the second half of the event, several of our guests held round-table discussions on the new paradigm of AI Agent

Guest:

Blast brightness: MuleRun CTO

Box: Monad Foundation Development Relations

Christian: Infini Founder

Professor Xu Jialong: Associate Vice-Rector (Academic Development), Hong Kong University of Science and Technology

"Agent's moat, does it exist?"

This issue has brought the four guests to the same side。

The first comment was made by Box: "Most of Agent's moats are weak, and Claude has a single employee who can reverse what you did in three weeks."

He gave an example: last year a bunch of start-ups were working as Co-work, Claude, and these companies collapsed。

"Whether it's faster than renewal or more influential than the founders." Or wait for death."

The beams are technically complementary: model gaps are accelerating, and the national production model is fast approaching GPT-4.6. Coding is more explosive, with two people doing one function for a month and three days alone。

"The competition function? A week is enough."

Then what's the moat

The beam highlights two points:

First, exclusive data. There are sensitive data that only you can get, and no one else can。

Second, user memory. Gemini doesn't provide memory export, you can't move out long enough, and that's really sticky。

2026, how complicated can Agent make a deal

The answer from Box is very short: "All human beings can do, all Agent can do. @Lego Group, whatever. But he said, "I don't want to let AI make money for you."

You drive 100 Agents at the same time, one always runs out, but that's probability, not strategy。

"THE PEOPLE WHO CAN MAKE MONEY ARE THOSE PEOPLE. AI CHANGES SPEED, DOESN'T CHANGE POWER."

Christian has made a more specific judgement: on the information level, the chain of sight, the event-driven strategy has been achieved, Agent can run 7x24 hours, the financial efficiency level, lightning, leverage wind control, liquidity optimization, and humans cannot monitor continuously, Agent can。

"It's not possible to stare at the fluidity of the loan agreement every day, but Agent can. It's not possible for a person to stare at a leverage coefficient while he's sleeping, but Agent can

People want to carry out a trade strategy that Agent can almost do, but if you can make money, you can't make money。

"Agent, how will he change the financial system?"

Professor Xu began by drawing a line between "value added" and "speculation"。

There is value in analysing company assets and finding the best analytical methods. Using Agent to predict market bets is no different from the traditional world。

From the bright point of view of the beam: Agent would level the information gap。

I used to make money because there were beles in the market that were better than you, and now, aided by AI, the new entrants are growing rapidly. Those who didn't trade before can now write scripts, do backshows, run strategies。

"The old man needs to re-establish his strength, or else the cuisine is yourself."

"Stabilize coins and Agent, what sparks?"

"STABLE CURRENCY IS A NEW INFRASTRUCTURE, AND NATURALLY CLOSER TO AI."

Box, he gave the example of take-out, and Agent is already able to order you some milk tea, and it'll be much bigger if you get access to the poaching room. There would be very strong synergies if stabilization coins were issued in enterprises with physical landing scenarios。

Christian was a pragmaticist, and he said that there were two points in the nature of the stable currency: a replacement dollar + a tool in a low financial system。

"It will take at least a month or two for traditional companies to start collecting money. A collection link is now generated in one minute, sent to the client, and paid upon request”

A hundred times faster, that's what Agent+Stabilizes most。

Professor Xu goes back to the bottom logic: the stability currency provides transparency and non-removable records, and it is popular in low-trust areas because it provides safeguards. “But platform players may hate the stabilization currency the most, exposing the funds allocated behind them.”

He concluded by adding that the aim of technology was to improve social values, not zero-sum money, which was too low if it was just to make money。

Skills Demo Presentation

Agency Wallet / Brad Bao

The Agenic Wallet, published by Cobo, is an Agent wallet with capacity for consumer control, approval and full process auditing. Its core idea is to sign a Pact contract for your Agent instead of delivering a private key。

Brad showed three scenarios from simple to complex. The simplest Uniswap deal, Agent submits the contract, and humans approve it on the phone。

A more complex cross-chain transfer, Agent himself charted the fastest and lowest rates, failed himself, Debug, and did not need human intervention。

The most complex Hyperliquid hedge strategy, Agent has developed its own arbitrage formula, and when the contract enters into force, no additional authorization is required for all operations in the next 30 days. The bottom of this product is that Cobo has been doing MPC multi-party security computing for years, and the money is always in your hands, and Agent can run, but can never take control。

"Trust is not an additional function on the product, but the foundation of the Agent economic system. "Brad says。

Macro surveillance

xingpt has been investing in Crypto VC for many years and has recently moved to AI content and investment studies. His core point: the tools are enough, and what's missing is how to use them。

He built a community called Web Trading to share Agent trading strategies. He's looking mainly for Beta, not Alpha。

Agent:

The first is geopolitical risk surveillance, Agent, dedicated to the Strait of Hormuz. The daily capture of ship traffic and traffic, compared to first-hand sources in Iran and the United States, is much faster than second-hand and third-hand news from the national media. He made 15 per cent of his three days of oil from this surveillance。

The second is Agent, a combination of data, technical indicators and social sentiments, which sends a buy-in signal when a multi-tempo resonates, an indicator that was resonated once at 60,000 bitcoons and validated its validity。

And the third is American equity, Agent, who has responded: US equity has had less than 20 percent of the largest retreat in the last decade, Sharp's rate is excellent and suitable for long-term configuration。

He then made two recommendations: first, that the database was important; and second, that the exchange of experts under the line was more important。

"You need to talk to industry experts once a week."

Active: Donut / Chris Zhu (founder of Donut)

Donut, located in the "active trading" scenario of high-risk assets, does not wait for a user's order, but takes the initiative to push trading opportunities to the user。

The product Demo shows a complete Agent assessment framework. The system will automate every answer from Agent, every data source, every tool call: time-consuming, pass rate, user emotional feedback, etc. This framework ensures that Agent is able to continuously optimize under a data drive。

The current product covers two categories of users:

• "Professional users" using depth features such as backsight, point analysis, simulation disks, etc. on the Dashboard

• Direct use of Telegram Bot by “white users”, with bottom models and data sources fully consistent with professional versions and non-differentiated

Chris Zhu shares three product methodologies:

"Product should be thick enough" -- not just an API or Bot. UI/UX is the key vehicle that allows users to understand what Agent is doing. Agent ' s decision-making process must be interpretable and visualized, otherwise trust cannot be built。

"Data feedback closed fast enough" -- every time a user puts down a list, every PNL, it's a feedback signal. The data points must return to the Evaluation system in real time and tell the model: "You took this proposal and the result was profit/loss. The more the feedback is closed, the less the delay, the faster the Agent evolves。

"Teams need to be systematic enough" - each team member should have his or her own Agent, automatically summarize the key messages that everyone is looking at every day and channel the context to other members as a central nerve. This way greatly increases the team speed。

The Donut team successfully landed on the "systemic team" model by developing the active hub AI Agent "Turing Bot", breaking down the tools limitations of passive commands, breaking the island by sending to members a daily summary of the exclusive context, active route information, and automatically settling real-time updates to the structured knowledge base, making it a central node for the active delivery of information, eliminating the team message blind zone and efficiently linking the overall context。

Vertical AI Employee:Agenda / Scarlett

The Agese team does not do generic Agent, chooses to do vertical AI Employee. A unique AI for each post to solve the problem of a particular scene。

They're already running four products:

The first is Meeting Copilot. Real-time multilingual translation, automatic summary of the proceedings and action Items, which are embedded in the customer service system, live webcasts and online classrooms。

The second is the job-seeking assistant, Pink Potato. After uploading the resume, the system automatically screens LinkedIn, Boss's direct high-scoring job, pushes it to a cell phone at less than a cent for each match, and Agent can even talk about your salary。

The third is Security Auditor CodeAuth. Upload code, scan security gaps free of charge, currently covering EVM and Solana ecology。

The fourth is Life Cloud, a life assistant, who orders restaurants and meals through Telegram or Line, and is currently working on the Southeast Asian market。

Forecast Market / Ryan

Ryan considers the projected market to be the most appropriate asset class for Agent because of its fragmentation, fragmentation and event-driven nature. Over the past few months, Agent has made millions of dollars on it。

Insider Bot shows three examples。

First of all, there's the smart return system. How can you stop the loss with a wallet? How can you run better than the target purse? Agent can help you optimize。

The second is smart trading systems. After access, Agent can access all the smart money databases, see the signal, and the user can ask it in its natural language, "Why the deal."。

Third, Tg Bot, the user can execute the transaction with a word。

Ryan showed their wallet database on the spot, and all indicators were updated in real time, not only as a demonstration of technology, but also as a precursor to the future capital shape of Agent transactions。

The Demo display is over, but the development of AI Agent is just beginning. Agent has come out of the lab, and they're writing tactics, chains, cross-chains, talking about pay, closing holes. The question is no longer "can" but "who's responsible" and "how big."。

AI Agent+ financial end of chain is a "trust system"

Underline activities under the line "Depreciating Web 4.0: When AI Agent takes over the chain of authority" ended in Hong Kong. Web3 for integration of MPC wallets, social transfers and chain security application NOXCAT to attend the event and disclose its chain-guaranteed contractual mechanism - - When users engage in off-site transactions, the funds are automatically locked by smart contracts, which are confirmed by both parties before being released, eliminating the risk of unilateral runaways at the institutional level。

NOXCAT states that this function is designed to address the pains of the long-term lack of trust guarantees in Web3 and is expected to be officially online in June 2026。

This event, which lasted several hours, finally focused on one word: trust。

Technology is just a tool; data is too much, and it's just a raw material. What really makes Web 4.0 run is a "trust mechanism" that everyone can trust. The future financial world is likely to be full of AI Agent, helping individuals, companies and even agreements to make decisions, trades, collaborates, and outpace human beings。

Instead of looking at the details of the execution, human beings are going to set the rules, give the strategy, target the results, and move from "beat the workers" to "make the rules of the game"。

And the future is not that far。