DeAI will define the next trend of Web3 artificial intelligence.
Written by: 0xJeff Source
Compiled by: Shan Ouba, Golden Finance
Since Trump took office, cryptocurrency investments have become exceptionally difficult due to broader uncertainties pushing capital into safe-haven assets.
The whole world is watching how the tariff situation is getting worse. Cryptocurrencies are no exception - BTC has already shown signs of strengthening, while Fartcoin has performed even stronger, outperforming all other assets.
Everything else is struggling
But aside from these two types of assets, everything else (I mean literally everything) is struggling — the cryptocurrency AI sector, which once held a dominant position, has significantly declined, with a total market value hovering around $6 billion. The situation for DeFi isn't much better, as capital has fled to other safe assets outside of cryptocurrencies, causing on-chain TVL to evaporate by over $50 billion.
So, what should we invest in?
This raises a question: how and what should we invest in during a volatile market?
Most of the people I know might point to yield farming on Berachain, Sonic, etc.—and that's fine. But for me, there are more interesting opportunities to explore that offer a better risk-reward ratio, especially in times of crisis.
In my opinion, the most asymmetric bet at the moment lies at the intersection of DeAI infrastructure and AI agents (which will be detailed later).
Adhere to this motto: "Be fearful when others are greedy, and be greedy when others are fearful."
The Subfields of Cryptocurrency and Artificial Intelligence That I Follow
In my opinion, there are several particularly interesting subfields in the field of crypto artificial intelligence at present.
Development Tools - Frameworks, Vibe Coding Tool, MCP Infrastructure
Consumer Artificial Intelligence - AI Agents, Alpha Tools, Games, DeFAI, GambleFAI, Personal / Companion (this does not include all subfields, but you get the idea.)
Framework Trends
To gain a more detailed understanding of consumer artificial intelligence / AI agents and development tools, I created this post in March (initially planned to do such posts monthly, but it seems that the progress in the general agent market is not sufficient for monthly updates):
frame
The FDV valuation during the last season (from October to November last year) was quite high, but as developers realized that many things could not be accomplished using off-the-shelf frameworks, and that LLMs might not be the best choice for financial use cases (as they are vulnerable to prompt injection attacks), the demand for these frameworks diminished.
Nevertheless, we still see growth in open-source frameworks and tools, such as @elizaOS (with 15.5k stars on GitHub), @arcdotfun (3.4k stars), and @sendaifun (1.2k stars), which received 434 stars, 197 stars, and 110 stars respectively last month.
Why Agent Distribution Network > Framework
I personally feel that frameworks are not that exciting because they do not have much value accumulation. Investment distribution networks / proxy centers are much better because there is clear value accumulation — that is, the transaction fees from the trading volume of speculators/investors trading AI proxy tokens. @virtuals_io is still a leader in this regard. Even if the daily trading volume drops from 8-9 figures to 7 figures, Virtuals remains the most trusted ecosystem by developers and is the most diverse ecosystem with many teams trying to build unique proxy products.
@elizaOS is starting to look more interesting, especially after @autodotfun (their launch platform) just went live. The team now has a distribution network that can directly accumulate value back to the $ai16z token.
What they need to address are the execution issues of high-quality partner project launches, so that they can meaningfully differentiate from the services provided by Virtuals (otherwise they will remain stuck in low-quality junk projects with market capitalizations in the 4-5 digit range).
Regardless, to take a step back, while these AI agents, frameworks, and distribution networks are interesting, the area with the best risk-reward ratio at the moment is decentralized AI infrastructure.
Why?
If you have been working in the field of artificial intelligence agents for a while, you may have noticed that the progress of agent products is roughly as follows:
Entertainment Dialogue "Agent" ➔ Alpha Analysis / Tool Dialogue "Agent" ➔ Trading Agent ➔ DeFAI Abstraction Layer ➔ Other Smaller Narratives ➔ Agents with Smarter Contexts, Multi-Agent / Groups, etc.
Death Wheel Trap
The reason many teams get stuck is that there are no proper core artificial intelligence products among these "proxy products." The only AI is the automatic prompting of the LLM to produce incessant chatter every x time.
Clearly, there have been significant changes compared to earlier times, but the reliance on LLMs or ready-made frameworks/workflows remains the same. Therefore, with every advancement and narrative of agent products, secondary agent products are created without proper use cases. (Similar to teams that forked major DeFi protocols a year later and gradually disappeared.)
This has led many teams to create hype through their proxies and token minting, but subsequently failing to maintain that attention (because there is no actual product), resulting in a death spiral (declining attention, declining token price).
Proxies need infrastructure; infrastructure builders need proxies.
However, while these teams may fail, they excel at one thing - that is GTM (Go-To-Market Strategy) / creating hype.
If there are many teams skilled in proxy GTM, knowing how to play the token game / build communities, but lacking appropriate AI products - what should they do? They should leverage the expertise of AI models and machine learning capabilities from inference networks and DeAI infrastructure providers.
On the other hand, the DeAI infrastructure team is not good at GTM. They are not on the front lines, and some of them are not crypto-native and do not know how to build a community.
So... why not combine the two?
I believe that the missing link between deep artificial intelligence infrastructure and viral agent distribution is where the real opportunity lies.
My Cryptocurrency AI Investment Theory
This leads to my theory of investing in crypto artificial intelligence:
Investing in DeAI infrastructure and introducing a new, unique Web3 workflow team that changes the way people interact with existing crypto products (DeFi, on-chain).
In Web2, workflow automation and enhancement—improving productivity while minimizing costs (thereby increasing profits)—is very common in the vertical agency field, especially for mundane tasks (the more mundane, the higher the value). For example:
Legal AI agents ingest original paper documents, create a legal case database, and collaborate with lawyers to help their clients achieve success in court.
Accountants review receipts, invoices, general ledgers, and trial balance sheets, and generate unaudited financial statements and tax returns.
Building agents review architectural blueprints, estimate costs, and propose methods to reduce construction/material costs while maintaining durability and aligning with client needs.
There are many case studies like this in Web2, where these startups rapidly grew to 7-8 figure ARR (Annual Recurring Revenue) within a few months - they truly use AI agents to automate and enhance workflows, providing real value to other businesses/customers.
In Web3, this is still quite novel and complex. To truly enhance workflows in DeFi, you need domain expertise. You need to understand the pain points that DeFi users (and regular users) face—and how to improve them. The DeFAI abstraction layer addresses this issue to some extent, but most are still unusable, with poor reasoning capabilities (you have to prompt very specific prompts to make it work—this actually backfires because ideally, you want regular users to use it, and regular users often don't know what they want to do, so they naturally don't know what to prompt).
This is why I believe that teams capable of meaningfully changing Web3/cryptocurrency workflows are very rare. However, if you can find them and invest in them early (now), you will have a lot of upside potential in the future.
On the other hand, we have the DeAI infrastructure. Most of it is not investable due to still being in the early stages.
These teams tend to raise millions of dollars from venture capital and require several years to conduct the TGE (Token Generation Event). Some projects that have already launched have experienced price declines of 50-80% due to market conditions. Those projects that perform well need to generate substantial revenue to maintain the token price (or hire a very good market maker).
@getgrass_io is a great example – reportedly generating 8-9 figures in revenue and being an excellent product for consumers (anyone can contribute bandwidth to receive airdrops).
Projects like Grass are very rare in the venture capital-backed DeAI infrastructure, and typically the only way to get involved early is to use the product / participate in an airdrop. They are likely to inflate the token price at TGE (low circulation, high FDV style), as venture capital enters at relatively low valuations. If you decide to invest in similar projects, the likelihood of losing money is higher than that of making money.
Investable Community Priority DeAI Ecosystem
This leads to another option - a pure community / no venture capital DeAI ecosystem. Yes, that is Bittensor.
Before the dTAO upgrade, the ecosystem was quite dull. Validators acted as some sort of capital allocators, as they decided which subnet received $TAO emissions (capital).
But since the upgrade and launch of dTAO on this year's Valentine's Day, there has been a huge change in this dynamic. Now the market determines which subnet receives emissions. The community—people—are now the capital allocators. If the community believes your subnet has no product and does not provide much value, you will not receive emissions (capital). This encourages subnets to build publicly, release faster, and create products that people truly want.
@BarrySilbert is betting on the Bittensor ecosystem through @YumaGroup (a subsidiary of DCG), which invests in, builds, and incubates Bittensor subnetworks. A recent interview with @RaoulGMI and @BarrySilbert has generated a lot of excitement in the community (as a major cryptocurrency institution has now entered the Bittensor ecosystem):
From an investment perspective, the liquidity of the Bittensor ecosystem is much better than that of the artificial intelligence agent ecosystem. The core issue with agent ecosystems like Virtuals is the pairing of LPs with Virtuals, which results in liquidity providers facing higher volatility and more impermanent loss.
This is why liquidity is often very low - you can typically only deploy between $1,000 to $5,000 and experience a slippage of 3-7% on these proxy tokens. On the other hand, deploying a similar amount into subnet tokens will only result in slippage of 0.05%-0.1% (or even lower).
Quick Summary:
The hype cycle of crypto AI agents is fading, and real products + user retention are still rare.
DeAI infrastructure is underestimated, misunderstood, and mispriced.
The best strategy is to combine infrastructure + proxy GTM to unlock new workflows.
$VIRTUAL leads the agency metaverse, Bittensor leads the infrastructure metaverse
Pay attention to the teams merging the two - if discovered early, there will be significant upside potential.
Summary
I believe DeAI will define the next trend of Web3 artificial intelligence. We will see more teams changing the way we interact with each other and with protocols, changing the way value is created, and generating new areas that reach more users and capture more market share (more mainstream). Now is the time to quickly understand the DeAI infrastructure and how it is changing things. Be sure to keep a close eye on teams that can successfully combine DeAI and agents.
Please remember that my theory is not set in stone. I have been constantly learning and refining it. I am doing my best to ensure that we can capture the next major trend in Web3 artificial intelligence. Again, this is not financial advice—please do your own research and take a cautious stance towards everything mentioned in this article.
The content is for reference only, not a solicitation or offer. No investment, tax, or legal advice provided. See Disclaimer for more risks disclosure.
The next opportunity for encryption AI: the integration of infrastructure and agents
Written by: 0xJeff Source
Compiled by: Shan Ouba, Golden Finance
Since Trump took office, cryptocurrency investments have become exceptionally difficult due to broader uncertainties pushing capital into safe-haven assets.
The whole world is watching how the tariff situation is getting worse. Cryptocurrencies are no exception - BTC has already shown signs of strengthening, while Fartcoin has performed even stronger, outperforming all other assets.
Everything else is struggling
But aside from these two types of assets, everything else (I mean literally everything) is struggling — the cryptocurrency AI sector, which once held a dominant position, has significantly declined, with a total market value hovering around $6 billion. The situation for DeFi isn't much better, as capital has fled to other safe assets outside of cryptocurrencies, causing on-chain TVL to evaporate by over $50 billion.
So, what should we invest in?
This raises a question: how and what should we invest in during a volatile market?
Most of the people I know might point to yield farming on Berachain, Sonic, etc.—and that's fine. But for me, there are more interesting opportunities to explore that offer a better risk-reward ratio, especially in times of crisis.
In my opinion, the most asymmetric bet at the moment lies at the intersection of DeAI infrastructure and AI agents (which will be detailed later).
Adhere to this motto: "Be fearful when others are greedy, and be greedy when others are fearful."
The Subfields of Cryptocurrency and Artificial Intelligence That I Follow
In my opinion, there are several particularly interesting subfields in the field of crypto artificial intelligence at present.
Framework Trends
To gain a more detailed understanding of consumer artificial intelligence / AI agents and development tools, I created this post in March (initially planned to do such posts monthly, but it seems that the progress in the general agent market is not sufficient for monthly updates):
The FDV valuation during the last season (from October to November last year) was quite high, but as developers realized that many things could not be accomplished using off-the-shelf frameworks, and that LLMs might not be the best choice for financial use cases (as they are vulnerable to prompt injection attacks), the demand for these frameworks diminished.
Nevertheless, we still see growth in open-source frameworks and tools, such as @elizaOS (with 15.5k stars on GitHub), @arcdotfun (3.4k stars), and @sendaifun (1.2k stars), which received 434 stars, 197 stars, and 110 stars respectively last month.
Why Agent Distribution Network > Framework
I personally feel that frameworks are not that exciting because they do not have much value accumulation. Investment distribution networks / proxy centers are much better because there is clear value accumulation — that is, the transaction fees from the trading volume of speculators/investors trading AI proxy tokens. @virtuals_io is still a leader in this regard. Even if the daily trading volume drops from 8-9 figures to 7 figures, Virtuals remains the most trusted ecosystem by developers and is the most diverse ecosystem with many teams trying to build unique proxy products.
@elizaOS is starting to look more interesting, especially after @autodotfun (their launch platform) just went live. The team now has a distribution network that can directly accumulate value back to the $ai16z token.
What they need to address are the execution issues of high-quality partner project launches, so that they can meaningfully differentiate from the services provided by Virtuals (otherwise they will remain stuck in low-quality junk projects with market capitalizations in the 4-5 digit range).
Regardless, to take a step back, while these AI agents, frameworks, and distribution networks are interesting, the area with the best risk-reward ratio at the moment is decentralized AI infrastructure.
Why?
If you have been working in the field of artificial intelligence agents for a while, you may have noticed that the progress of agent products is roughly as follows:
Entertainment Dialogue "Agent" ➔ Alpha Analysis / Tool Dialogue "Agent" ➔ Trading Agent ➔ DeFAI Abstraction Layer ➔ Other Smaller Narratives ➔ Agents with Smarter Contexts, Multi-Agent / Groups, etc.
Death Wheel Trap
The reason many teams get stuck is that there are no proper core artificial intelligence products among these "proxy products." The only AI is the automatic prompting of the LLM to produce incessant chatter every x time.
Clearly, there have been significant changes compared to earlier times, but the reliance on LLMs or ready-made frameworks/workflows remains the same. Therefore, with every advancement and narrative of agent products, secondary agent products are created without proper use cases. (Similar to teams that forked major DeFi protocols a year later and gradually disappeared.)
This has led many teams to create hype through their proxies and token minting, but subsequently failing to maintain that attention (because there is no actual product), resulting in a death spiral (declining attention, declining token price).
Proxies need infrastructure; infrastructure builders need proxies.
However, while these teams may fail, they excel at one thing - that is GTM (Go-To-Market Strategy) / creating hype.
If there are many teams skilled in proxy GTM, knowing how to play the token game / build communities, but lacking appropriate AI products - what should they do? They should leverage the expertise of AI models and machine learning capabilities from inference networks and DeAI infrastructure providers.
On the other hand, the DeAI infrastructure team is not good at GTM. They are not on the front lines, and some of them are not crypto-native and do not know how to build a community.
So... why not combine the two?
I believe that the missing link between deep artificial intelligence infrastructure and viral agent distribution is where the real opportunity lies.
My Cryptocurrency AI Investment Theory
This leads to my theory of investing in crypto artificial intelligence:
Investing in DeAI infrastructure and introducing a new, unique Web3 workflow team that changes the way people interact with existing crypto products (DeFi, on-chain).
In Web2, workflow automation and enhancement—improving productivity while minimizing costs (thereby increasing profits)—is very common in the vertical agency field, especially for mundane tasks (the more mundane, the higher the value). For example:
There are many case studies like this in Web2, where these startups rapidly grew to 7-8 figure ARR (Annual Recurring Revenue) within a few months - they truly use AI agents to automate and enhance workflows, providing real value to other businesses/customers.
In Web3, this is still quite novel and complex. To truly enhance workflows in DeFi, you need domain expertise. You need to understand the pain points that DeFi users (and regular users) face—and how to improve them. The DeFAI abstraction layer addresses this issue to some extent, but most are still unusable, with poor reasoning capabilities (you have to prompt very specific prompts to make it work—this actually backfires because ideally, you want regular users to use it, and regular users often don't know what they want to do, so they naturally don't know what to prompt).
This is why I believe that teams capable of meaningfully changing Web3/cryptocurrency workflows are very rare. However, if you can find them and invest in them early (now), you will have a lot of upside potential in the future.
On the other hand, we have the DeAI infrastructure. Most of it is not investable due to still being in the early stages.
These teams tend to raise millions of dollars from venture capital and require several years to conduct the TGE (Token Generation Event). Some projects that have already launched have experienced price declines of 50-80% due to market conditions. Those projects that perform well need to generate substantial revenue to maintain the token price (or hire a very good market maker).
@getgrass_io is a great example – reportedly generating 8-9 figures in revenue and being an excellent product for consumers (anyone can contribute bandwidth to receive airdrops).
Projects like Grass are very rare in the venture capital-backed DeAI infrastructure, and typically the only way to get involved early is to use the product / participate in an airdrop. They are likely to inflate the token price at TGE (low circulation, high FDV style), as venture capital enters at relatively low valuations. If you decide to invest in similar projects, the likelihood of losing money is higher than that of making money.
Investable Community Priority DeAI Ecosystem
This leads to another option - a pure community / no venture capital DeAI ecosystem. Yes, that is Bittensor.
Before the dTAO upgrade, the ecosystem was quite dull. Validators acted as some sort of capital allocators, as they decided which subnet received $TAO emissions (capital).
But since the upgrade and launch of dTAO on this year's Valentine's Day, there has been a huge change in this dynamic. Now the market determines which subnet receives emissions. The community—people—are now the capital allocators. If the community believes your subnet has no product and does not provide much value, you will not receive emissions (capital). This encourages subnets to build publicly, release faster, and create products that people truly want.
@BarrySilbert is betting on the Bittensor ecosystem through @YumaGroup (a subsidiary of DCG), which invests in, builds, and incubates Bittensor subnetworks. A recent interview with @RaoulGMI and @BarrySilbert has generated a lot of excitement in the community (as a major cryptocurrency institution has now entered the Bittensor ecosystem):
From an investment perspective, the liquidity of the Bittensor ecosystem is much better than that of the artificial intelligence agent ecosystem. The core issue with agent ecosystems like Virtuals is the pairing of LPs with Virtuals, which results in liquidity providers facing higher volatility and more impermanent loss.
This is why liquidity is often very low - you can typically only deploy between $1,000 to $5,000 and experience a slippage of 3-7% on these proxy tokens. On the other hand, deploying a similar amount into subnet tokens will only result in slippage of 0.05%-0.1% (or even lower).
Quick Summary:
Summary
I believe DeAI will define the next trend of Web3 artificial intelligence. We will see more teams changing the way we interact with each other and with protocols, changing the way value is created, and generating new areas that reach more users and capture more market share (more mainstream). Now is the time to quickly understand the DeAI infrastructure and how it is changing things. Be sure to keep a close eye on teams that can successfully combine DeAI and agents.
Please remember that my theory is not set in stone. I have been constantly learning and refining it. I am doing my best to ensure that we can capture the next major trend in Web3 artificial intelligence. Again, this is not financial advice—please do your own research and take a cautious stance towards everything mentioned in this article.