Original Title: 'Post-AI Agent Bubble: Where's the Real Value in Web3 AI?'
Author: 0xJeff
Translated by: TechFlow
Overview
AI agents rapidly grew to over $20 billion in just a few months, only to collapse just as quickly. However, the field is maturing. Infrastructure, decentralized AI, and practical applications are taking over. How the next wave is taking shape and why it's worth paying attention.
In the fourth quarter of last year, we witnessed the "AI agent" industry grow from zero to over $20 billion in just a few months - from interesting, charming, entertaining, and almost absurd "agents" to financial agents promising to change the world through trading and investment and make you incredibly wealthy. And not just agents to make you rich, but also many investment DAOs... DAOs that invest in other agents (or agents) DAO (3,3).
From Hype to Infrastructure
We all know that in a new industry (accompanied by new catalysts like Web2 AI, Trump's election supporting crypto and AI), people don't care about fundamentals. Anything that generates massive discussion, looks hyped, and has a cool-looking demo can easily surpass a $100 million market cap.
@virtuals_io became the ecosystem's leader, controlling market promotion, attracting builders' attention, telling the best stories, and weaving the best narratives. This attracted builders to publish projects on Virtuals and drew retail traders' attention for hype.
Then, @elizaOS emerged with a different approach - open-source AI, providing tools for developers wanting to create "gold mining" agents. A massive movement formed around this concept, with adoption growing at an astonishing speed, with GitHub stars and forks rapidly increasing (and still growing).
Virtuals' valuation grew to over $5 billion, with Eliza accounting for about half at its historical peak, and many other interesting agents reaching 8-9 digit peaks, such as AIXBT reaching $1 billion. Of course, the situation is now very different, with newly launched agents averaging $3-10 million in trading price. Well-performing older agents average $10-50 million. Valuation caps have compressed, with total market cap shrinking from $20 billion to the $4-6 billion range.
Infrastructure Momentum and Web2 Acceleration
The market now focuses on "pure fundamentals", with people more concerned about infrastructure and decentralized AI, especially as AI models in Web2 continue accelerating rapidly - Meta's Llama, OpenAI's GPT, Grok, DeepSeek, Alibaba's Qwen are releasing new improved and optimized models every month. You can see how ChatGPT's image generation model immediately created a viral "Ghibli-fication" trend upon release.
Moreover, with AI model capabilities improving, Web2's consumer level is advancing faster, making previously impossible things now possible - Lovable, Bolt, Cursor, Windsurf enable developers to release more products faster. Agent workflows and AI agents are ubiquitous. Entry barriers are lowered, and user conversion costs are almost zero. If you don't like an application, you can easily find a competitor service or product with more favorable pricing and better UI/UX.
Data Ownership Awakening
Meanwhile, many people started thinking: "If there are so many agent applications using centralized technology, who owns my data? Where will my data go? If I discuss something private with AI, will it remain confidential? Or will it flow elsewhere?" This is especially important because of OpenAI's recent update where ChatGPT's memory feature can reference all your past chat logs to provide more personalized responses.
Bro... this sounds cool and could potentially trigger a wave of personalized AI agents, co-pilots, private secretaries, therapists, partners, etc. You can imagine the implications when your data is owned or controlled by others.
The Rise of Decentralized AI (DeAI)
I made some predictions last year, one of which was that decentralized AI would emerge in Q2 2025, with infrastructure enhancing confidentiality, transparency, verifiability, and ownership, thus gaining more adoption and attention due to the growing demand for these characteristics.
There are three independent trend sectors, with many trends interweaving or converging between them:
Web2 AI VC Trends (YC companies launching vertical agents, a16z positioning future consumer trends through their arguments, Perplexity launching its AI fund)
Web3 AI VC Trends (DeAI infrastructure investment, distributed training, inference networks, etc.)
Web3 AI Retail Trends (AI agent ecosystem, consumer agents, AI consumer applications)
Web2 vs Web3 AI: Distinctly Different Energies
For Web2, due to the significantly larger total addressable market (TAM), many enterprises want to transform or optimize their businesses with AI, improve workflows to generate more leads, more conversions, more sales, retain more customers, reduce management costs, and operate at a higher level, with many businesses seeking solutions that solve highly specific pain points in their domains.
This optimization demand attracts many young startup founders seeking better ways to introduce AI agents to improve workflows. Compared to traditional SaaS, AI agent solutions can significantly save capital or generate more leads. This allows agent startups to charge higher subscription fees (which is why we see many startups reaching 7-8 digit annual recurring revenue in just a few months).
For Web3 VCs, the trends here are very different because blockchain provides the perfect layer for decentralized AI (DeAI), such as verifiable/immutable transaction trails, trustless environments, decentralized computing, trust-minimized AI inference and training. In short, the future direction is to let people know how their data is processed, understand AI's thought process, own their data, own models, own use cases, and be incentivized to share (without censorship), and so on. Web3 VCs have been investing in these futures.
Why Retail Loves Agents (Even If They Don't Fully Understand Decentralized AI)
For Web3 retail, decentralized AI (DeAI) is very difficult to understand because it requires learning many terms and understanding important content (sometimes feeling like an alien language). That's why retail tends to choose the easiest to understand - starting with "Web3 AI agents" that can chat, be funny, and do some entertaining activities.
As retail continues to stay in this industry, they gradually realize this is insufficient to create sustainable value for users (yes, many AI agents are useless and lack creativity). This realization (coupled with market instability) drives market consolidation, with useless agents gradually dying off while useful agents survive (though with significantly reduced valuations).
People start realizing the need for a core AI product with actual use cases. This realization motivates teams to either develop genuine AI products or collaborate with AI companies with real technology, such as @AlloraNetwork and @opentensor (Bittensor).
This shift has two benefits:
It helps people understand more about infrastructure they're less familiar with.
Provided actual use cases for AI agents to showcase to the community.
Before this transformation: Agents possessed basic skills/use cases (chatting, post analysis)
After this transformation: Agents possess advanced useful skills (AI-driven betting, trading, liquidity provision, mining, etc.)
Agents like @AskBillyBets and @thedkingdao become ideal representatives of the Bittensor subnet, bringing cool technology into the mainstream.
Bittensor Ecosystem
What I find interesting about the Bittensor ecosystem is that it's an ecosystem of decentralized AI that you can invest in. Today, most decentralized AI (DeAI) can only be invested in by venture capitalists or behind-the-scenes strategic investors, as it is still in the early stages, and many projects have not yet issued tokens.
But Bittensor allows anyone to hold their $TAO and stake it to the subnet they want to support, thereby converting it into the subnet's alpha token (investing in decentralized AI projects).
I have previously shared my fatigue with cross-chain and trading experiences, but the technology, products, and atmosphere there, especially @rayon_labs, are impressive.
I like Rayon Labs because they are building products that optimize consumer-facing UI/UX. Given the nature of dTAO, the market determines the emission and pricing of each subnet, so it becomes increasingly important for each subnet to build products that are easy to understand and master.
Rayon has many cool subnets (possibly the coolest being Gradients, an automated machine learning platform where you can easily train models), and even cooler is their latest flagship product, Squad AI agent platform, where you can create agents by dragging and dropping blocks (node builder style, similar to AI agent creation on @figma).
Final Notes
I am still in the early stages of my deep dive into Bittensor, and I will publish a dedicated article later, sharing interesting findings and how to leverage opportunities. If you want to learn about other trends and changes in the market, please check out this article.
Personal Note: Thank you for reading! If you are a Bittensor subnet owner or researcher, building or researching cool things in the Bittensor ecosystem, I would be happy to connect and learn more.
