Recently, SpaceX and Meta have successively launched AI agents, OpenAI and Anthropic models have reduced prices, blockchain payment has become a key element of the intelligent Internet, and AI Agent is reshaping the software economy and Web3.0.
Author: Meng Yan’s thoughts on blockchain
Some very interesting things have happened in the past month or so.
On August 11, SpaceX released the AI agent (agent) application Grok Bot. This is an "AI colleague" who works around the clock on his own cloud computer.
On September 8, Meta launched Muse, a personal intelligence application. It can open browsers, fill out forms, and bargain for users. It topped the list of the U.S. App Store within ten days of its launch, and triggered a round of selling of traditional online intermediary platform companies. For example, three online travel stocks, Expedia, Airbnb, and Booking, plummeted in consecutive days.
On September 22, Anthropic, which makes Claude, released a new model Opus 5.5. Not long ago, OpenAI, which makes ChatGPT, launched two new versions of GPT-6, with the price reduced to half of the previous generation.
In mid-September, Salesforce, one of the world's largest SaaS software companies, held its annual conference. Patrick Stokes, president of the application business, said that AI will take apart the software interface and then replace it.
These things come from different companies and industries, and they all happened within two months. They seem to be going from left to right, but underneath they are the same thing: AI agents that can do things for people start to deal with software and the Internet for people. The new generation of Internet is about to look like.
1. Road roller effect
The person in charge of an AI incubator told me that the biggest awakening for AI entrepreneurs in the past year is that they discovered that AI is different from the Internet and blockchain. It is an extremely centralized and extremely centralized arena. Start-ups have a very short survival period. They must be quick and sell, and they must not hesitate to find buyers and cash out before being crushed by giants. As for your big dreams, don’t even think about them.
This is the mentality of the grasshoppers in front of the steamroller.
The “steamroller effect” is a source of anxiety for many AI entrepreneurs. Every time the cutting-edge model is advanced, a group of startups will be crushed without knowing it. A few listed companies can at least use the plummeting stock price as a long-lasting cry, while hundreds or even thousands of small teams can instantly go back to zero without anyone noticing.
Look at the money first. According to statistics from PitchBook, an American venture capital data agency, global AI venture capital hit a record in the first half of 2026, with more than half going to OpenAI and Anthropic. It was like a banquet with thousands of tables, half of the dishes were served at the main table, and there were only two people sitting at the main table.
Let’s look at the specific company. Google has an AI note-taking product called NotebookLM. Users put files and web pages into it, and it turns the content into podcast-style audio explanations. It became popular online in 2024. Its head, Raiza Martin, left Google with two colleagues and built Huxe, an AI podcast app that generates daily audio briefings based on users' emails and calendars. Google chief scientist Jeff Dean invested in it.
On May 21 this year, Spotify, the world’s largest music streaming platform, made an update that included similar features. On May 22, Huxe announced it was closing.
The same goes for listed companies. In February, Anthropic released a set of plug-ins for legal, financial and other industries to Claude. Shares of Thomson Reuters, the parent company of Reuters, which makes its living selling legal, tax and accounting information, fell more than 15% on the day.
After Claude released the Opus 5.5 model, a large number of users posted on Twitter the beautiful videos they made with simple prompt words and cheered. The team that had previously worked hard in the field of video AI production for countless sleepless nights could only face the screen and cry silently.
Heaven and earth are unkind and regard all things as stupid dogs. Giants like OpenAI and Anthropic are not trying to make trouble with you. They are not hostile to you and have no intention of competing with you. They don’t even know you exist. They just go along with the trend, and then you will be wiped out.
How did you say that shitty line? Destroying you has nothing to do with you.
2. Workshop-oriented software economy
In February of this year, Claude Opus 4.6 was released. With the support of this model, Claude Code suddenly became so powerful that even people who couldn’t write code suddenly felt that their technical dreams could still be saved.
There are more and more people making software. There is an AI platform called Lovable, which allows people who cannot program to use chat to create websites and applications. It says that the platform adds 1 million new projects every week, and its users are mainly programmers who can't write even half a line of code.
On the other side, selling software has a hard time. Major index funds tracking U.S. software stocks fell more than 24% in the first quarter of this year, their worst quarter since the 2008 financial crisis. The stock price is based on expectations, and there are also factors such as high interest rates, which does not mean that demand has shrunk. In September, it was Internet intermediaries that were hit, and the market was re-pricing software sold for human use.
I call this the "workshop-ization" of the software economy. The Industrial Revolution moved spinning and weaving from households to factories, and AI is moving software writing from factories to households. Everyone and every small company can open a software workshop for themselves.
In the past, when opening a small store, you needed an inventory system. You had to find an outsourcing company to discuss requirements, sign a contract, and wait for the construction period. Now, just say a few words to the computer and you can quickly build a prototype in an afternoon. The same goes for internal tools such as customer management and personnel scheduling.
Could the next big company grow out of these workshops? It's hard. The problem of workshops is isolation, which becomes more obvious the more workshops there are. In a small country with few people, we can hear the sounds of chickens and dogs, and the people will not interact with each other until they grow old and die. I regard this as an ideal, but it is a dilemma when it comes to software: I can't use your product, and you can't touch my data. Things made in the workshop can only be used by yourself.
There have never been so many people making software, and it has never been so difficult to sell software.
3. Internet of Intelligent Agents
AI is extremely centralized, while software is becoming decentralized. There are two heavens of ice and fire, but they are actually two sides of the same body. It is precisely because your AI capabilities continue to strengthen that the software space is squeezed.
This is just the beginning. People are still using AI to develop software and then use the software. Going a few steps further, there is no need for software. Users directly interact with the AI Agent to solve all problems.
Muse is the clearest prototype yet.
Meta allocates a virtual computer to each user in the cloud, and Muse works on this computer. When the user closes the App, it continues to work, and notifies the user when it is finished or requires approval; if the other party has an interface, it calls it directly through the connector; if there is no interface, it opens the browser and operates page by page like a human.
Filling out forms, comparing prices, and negotiating prices are all tasks that users had to do themselves, and are left to it.
It's not done well yet. American payment media PYMNTS asked it to do three things: restock toilet paper on Amazon, order a Domino's pizza, and book a table at a restaurant. None of the three tasks were completed. The reviewer said that it added an extra layer of management to a task that could be completed by one person in 30 seconds.
But what we want to look at is the direction. How is this different from using apps in the past? In the past, to book a flight, you had to open several travel websites, compare prices one by one, and fill in your name and ID number in each box. I’ll tell Muse next month to find the cheapest direct flight to Tokyo and let it do the rest.
For the first time, the user only speaks to one Agent, which deals with the entire Internet. Apps and websites recede into the background.
This is not just a change in the way people interact with the Internet, it is another upgrade of the Internet itself.
Looking at the history of the Internet: Web 1.0 connects documents, the Internet of documents, for people to read; Web 2.0 connects applications and services, and it is the Internet of applications, for people to use. What does Web 3.0 connect?
The term Web 3.0 has had several terms over the past twenty years. Berners-Lee, the inventor of the World Wide Web, systematically elaborated on the "Semantic Web" in 2001, which requires labeling web pages with machine-readable labels. The blockchain circle has proposed the "Internet of Value", which wants money and assets to flow online like information. Both statements describe characteristics, but neither state the form.
It is now clear that Web 3.0 takes the form of the Internet of agents. Agents read, use, and talk for people. An Agent deals with hundreds of apps and thousands of APIs for people. Human experience is no longer important, only what the Agent thinks is good is really good, and then it’s good for me too. In the past two decades, product managers and designers have built products around human experience, focusing on interfaces and operations. In the future, the product will be written to the Agent, which requires a clear interface, clean data, and readable terms.
Salesforce launched a toolkit without an interface, opening the entire platform to Agent through MCP and API, and also installed its own functions directly into Claude. A company that relies on interfaces to sell software has dismantled the interface itself.
The CEO of online travel platform Expedia summarized the company's new strategy as "being present wherever agents are." Travel platforms used to pull people to their own websites, but now they have to pull people to Agents.
Who is this bad news for? Wall Street is repricing "consumer inertia." Many businesses rely on customers who are too lazy to compare prices, too lazy to change apps, or too lazy to call to negotiate. Agent takes no trouble and does all these things. So the market is selling off this business.
Even Meta itself needs to transform. This company started by selling users' attention, and its revenue mainly comes from advertising. Agent does not watch ads or browse information streams. The source of income Zuckerberg found for Muse is to draw a small fee from transactions.
Thirty years ago, website optimization for search engines began, and it later became a business. This time, the object that needs to be adapted is changed from a sorting machine to an Agent that makes decisions for the owner. Products that the Agent cannot understand do not exist; only products that the Agent can understand and use smoothly can win business. The same goes for the abilities made in the workshop. If the Agent can't understand them, they won't be sold.
US technology analyst Ben Thompson said agents will become "the ultimate gatekeepers". Whoever controls the Agent controls the user's needs.
Another thing that is bound to happen is blockchain payments and the token economy. The amount of value exchanged between Agents is small, the frequency is high, and the opponents are unfamiliar. Whether to pay and how much to pay are decided by the program on the spot, and no one looks at it. Card swipes charge 2.9% plus 30 cents per transaction, and paying 1 cent costs 30 times the fee, and the card network cannot handle these types of transactions.
Visa’s own research also admitted that cards cannot do this, and allocated micropayments between machines to stable coins. Stablecoins are digital currencies that are usually linked to the U.S. dollar on a one-to-one basis and circulate on the blockchain. The total amount has exceeded 300 billion U.S. dollars.
My judgment is that digital payment and token economy based on blockchain, that is, a set of arrangements that use on-chain certificates to calculate, settle and distribute income, are indispensable basic elements of the intelligent Internet.
I always remember that at a meeting at the Digital Asset Research Institute in 2018, teacher Zhu Jiaming said that in the long run, blockchain is not for people, but for AI. I thought this was right at the time, but I couldn’t figure out how to use AI. It's clear now.
4. “AI agent-friendly” work
The impact of AI on jobs has always been a hot topic. In the AI era, what kind of talents can survive and develop sustainably in the long term? Or to put it more bluntly, what job is safe from being crushed by a steamroller?
It must be a friendly collaboration with AI agents.
Of course there are top AI scientists in large model companies, but the number of such people is extremely small, the threshold is extremely high, and they are not accessible to most people.
Front-end deployment engineer (FDE), which was very popular some time ago, seems to have died out quickly. Everyone quickly discovered that this was just a fashionable term for on-site outsourcing, and most importantly, it was often a one-time event. I rushed to implement various imaginary AI enterprise solutions, but found that the data quality was too poor, and the AI deployed in the enterprise was like the second fool in the village. I had to reorganize it from the underlying data. But isn’t this what data engineering should do? As a result, FDE became Fooled Data Engineer, a fooled data engineer. Whoever loves this job can do it.
What is popular now are AI engineers who make products based on AI models and AI-Assistant Developers who use AI to assist in the development of traditional software. But if the two judgments that AI Agent will squeeze software and the Internet will move toward intelligent agents are correct, then these two types of jobs will also transform. In the past, we made products for people, but in the future, more and more products will be made for agents. The past was search engine optimization, and the future is agent optimization.
There is another kind of work that is more common, which is the management of AI agent organizations.
When an intelligent agent is powerful enough, you don't even need to develop software. As long as you can manage it well, make demands on it, and make judgments and decisions on key links, you can complete most of the work. But this is far from enough. In the future, you must be able to design and organize several, dozens, hundreds or even thousands of agents, organize them into efficient and powerful legions, design efficient processes, ensure security, control expenses and token budgets, and effectively evaluate performance and continuous improvement, and then compete and fight with your competitors' armies of agents of the same size.
For example, the purchasing agent of a company has to negotiate prices with the sales agents of a thousand suppliers at the same time, and the agents on the other side are equally smart. The outcome depends on how this side divides labor, authorizes authority, and holds people accountable. Management must become a project, and permissions, boundaries, incentives, and audits must be trainable and measurable. This is a new management science and a new system engineering.
Conway's Law states that an organization that designs a system will eventually create a system that replicates its own communication structure. In the Agent era, the members of the organization are Agents, and the system and organization are integrated. The Agent system is what the organization is like, and designing the Agent system is designing the organization.
This matter is not someone’s fantasy. Microsoft said in its annual workplace trends report that everyone will become the boss of Agent; NVIDIA CEO Jensen Huang said that the IT department will become the HR department of AI Agent.
Therefore, the most valuable jobs in the Internet of Agents are, first, AI engineers who can make products and services for Agents, and second, army leaders who can command a thousand Agents to defeat another thousand Agents.