Original title:Post-AI Agent Bubble: Where the Real Value in Web3 AI?

Original author: 0x Jeff

Original translation: Ethan,

Brief Overview:

The total market value of AI agents grew from 0 to over US$20 billion in a few months, and then quickly collapsed. But this field is gradually maturing. Infrastructure, decentralized AI and real practicality are gradually taking over. How will the next wave of development shape the future and why is it worth paying attention to?

In the fourth quarter of last year, we saw an area where the fastest growing “AI agent” grew from zero to over $20 billion in just a few months – from some fun, charismatic, entertaining “agents” to financial agents that promise to change the world and get rich overnight through transactions and investments. And...not just those agents that can make you rich, investment DAO has also emerged... Human (or agent) DAO (3, 3) invest in other agents.

From Hype to Infrastructure: The Evolution of Web3 AI Agents

We all understand that in an emerging field (and in the context of Web2 AI, Trump’s election and support for new catalysts such as cryptocurrencies and AI), people don’t care about fundamentals. Any demonstration that can make noise, looks hype and cool can almost quickly reach a market value of over $100 million.

@virtuals_ioIt has become an ecosystem that has occupied the market, seized market attention, told good stories, and created the best narrative. This has attracted a large number of creators to publish content on Virtuals and launch projects, while attracting the attention of retail investors to capitalize on hype.

@elizaosThen it appeared and a different approach was adopted - open source AI, allowing any developer to use a "shovel" to dig out gold. A huge spread effect was formed around this concept, adopting rapid growth, and the number of stars and branches on GitHub also increased dramatically (these numbers are still growing).

Virtuals’ valuation has grown to more than $5 billion, while Eliza accounts for about half of the market cap at its highest point, and some other interesting agents are also reaching 8-9-digit valuations, such as AIXBT, which hit $1 billion. Of course, the situation is very different now, with the valuation of newly released, well-performing agents ranging from $3 million to $10 million on average. Old well-performing agents have averaged between $10 million and $50 million. The upper valuation limit was compressed, and the overall market value also narrowed from US$20 billion to US$4 billion to US$6 billion.

Infrastructure acceleration and rapid advancement of Web2

The market is now beginning to focus on "pure fundamentals", and people are more concerned about infrastructure and decentralized AI, especially the AI ​​models in Web2 continue to accelerate at an astonishing speed - Meta's Llama, OpenAI's GPT, Grok, DeepSeek, Alibaba Qwen and others will release new improved and more optimized models every month. ChatGPT’s image generation model quickly triggered a “giplier” trend after its release.

Above all this, the consumer layer of Web2 has developed much faster than before due to the improvement of AI model capabilities - things that were impossible to achieve before have now become possible. Tools such as Lovable, Bolt, Cursor and Windsurf allow developers to launch more products faster. Agent workflow and AI agents are everywhere. The entry barrier is lowered, and the switching cost of users is almost zero - if you hate an application, you can easily find more competitive services or products, with a better interface and user experience.

Data ownership awakening: the call for decentralized AI

At the same time, many people began to think: "Since so many proxy applications are using centralized technology, who owns my data? Where will my data go? If I discuss some private content with AI, will it keep it secret? Or will it leak?" This question is particularly important, especially since OpenAI's recent update mentioned that ChatGPT is now able to quote your past chat history to provide a more personalized response.

Alas... all of this sounds cool and is likely to trigger a wave of personalized AI agents, such as co-pilot, personal secretary, therapist, partner, etc. But you can imagine what the consequences will be when someone has or has mastered your data.

Decentralized AI (DeAI): The Power to Lead the Future

I made some predictions last year, one of which is that decentralized AI will make its mark in the second quarter of 2025, and infrastructure will enhance confidentiality, transparency, verifiability and data ownership, and therefore gain more adoption and attention.

This trend can be divided into three main parts, many of which intersect or intertwined:

• Web2 AI Venture Capital Trends (YC Company launches vertical agents, a16z makes layouts for future consumption trends, and Perplexity launches AI Fund)

• Web3 AI Venture Capital Trends (DeAI Infrastructure Investment, Distributed Training, Inference Networks, etc.)

• Web3 AI retail trends (AI agent ecosystem, consumer agent, AI consumer applications)

Web2 and Web3 AI: The Collision between Two Worlds

For Web2, Since the total address market (TAM) is much larger than Web3, many companies are seeking to transform and optimize their businesses through AI and improve workflows to generate more leads, more conversions, higher sales, retain more customers, reduce administrative costs, and operate in a more efficient way. Therefore, many businesses seek solutions that can solve their specific pain points.

This need for optimization has attracted many young entrepreneurs to find better ways to bring AI agents into their workflow. Compared with traditional SaaS, the solutions provided by AI agents can save a lot of money or generate more leads. This allows agency startups to charge higher subscription fees for their use (which is why we see many startups reach 7-8-digit annual revenue in a few months).

For Web3 venture capital, the trends here are very different, because blockchain provides DeAI with a perfect infrastructure layer, such as: verifiable/untamperable transaction records, trustless environments, decentralized computing, trust-minimized AI reasoning and training (sorry, a lot of terms are used, but you should be able to understand what I want to express).

In short, the future development direction is for people to understand how their data is processed, understand the AI ​​thinking process, have their own data, have their own models and usage scenarios, and have the motivation to share (not censored), etc. Web3 venture capital has already invested in these future trends.

AI Agent Boom in the Retail Market: Not Just Entertainment

DeAI is very difficult to understand for the Web3 retail market because it requires you to learn a lot of terms and understand the key points (sometimes it feels like an alien language). This is why retail market users prefer the easiest things to understand—such as the "Web3 AI Agents" that start with conversational robots, which are funny and can do entertaining content.

As the retail market continues to penetrate the industry, they gradually realize that these basic skills of chat and analytics are not enough to create sustainable value for users. This perception (plus the bad market environment) prompts the market to integrate, useless agents gradually disappear, while useful agents still survive (although their valuations are greatly reduced).

People are beginning to realize that AI products must have a core practical use scenario. This awareness prompted the team to either develop real AI products or cooperate with truly technical AI companies, such as@Alloranetwork、@opentensor（Ask

There are two benefits to this transformation:

(i) It gives people a better understanding of infrastructure.

(ii) It provides real use scenarios for AI agents, allowing them to show their value to the community.

• Before the transformation:Agents with basic skills/use scenarios (chat, analysis, etc.)

• After the transformation:Agents with advanced practical skills (such as AI-driven betting, trading, liquidity provision, agriculture, etc.)

picture@AskbillyBets、@thedkingdaoSuch an agent has become an ideal agent, showing the Bittensor subnet, bringing cool technology to the mainstream.

Bittensor Ecosystem: New Opportunities for Decentralized AI

I think one of the interesting things about the Bittensor ecosystem is that it is an ecosystem full of decentralized AI that anyone can invest in. Today, most decentralized AI projects are limited to VCs or strategic investors participating in behind closed doors, as they are still in their early stages and many projects have not yet issued tokens.

But Bittensor allows anyone to stake their $TAO into the subnet they want to support, thereby converting it into the subnet’s alpha token (directly participating in the DeAI project).

Although I have publicly expressed my disappointment with the bridging and transaction experience, Bittensor’s technology, products and atmosphere are excellent, especially@rayon_labs  team.

I like Rayon Labs because they do a lot of consumer-friendly work in optimizing UI/UX. Given the characteristics of dTAO – the market determines the emissions of each subnet and the pricing of the subnet – it is therefore especially important for each subnet to build products that are easy to understand and understand.

Rayon has many cool subnets (the coolest of which is probably Gradients, an AutoML platform that can easily train models on the platform), and even cooler is their latest flagship product, Squad AI agent platform, which allows users to create agents through drag and drop boxes (similar to Figma’s AI agent creation method).

Conclusion

I'm still in the early stages of getting to know Bittensor in depth, and will be posting a dedicated article later to share interesting content I've found and show how to seize opportunities from it.