The next trend of crypto AI: the integration of infrastructure and agents

Author: 0xJeff Source: X, @Defi0xJeff Translation: Shan Ouba, Golden Finance

Since Trump took office, cryptocurrency investment has become extremely difficult as broader uncertainty pushes capital into safe-haven assets.

The world is watching how the tariff situation is getting worse. Cryptocurrency is no exception - BTC has shown signs of strength, while Fartcoin has performed even stronger, outperforming all other assets.

Everything else is struggling

But apart from these two assets, everything else (and I mean literally everything) is struggling - the once-prime crypto AI sector has fallen sharply, with the total market value remaining at around $6 billion. The situation in DeFi is not much better, with more than $50 billion in on-chain TVL evaporated as capital fled to other safe assets besides cryptocurrencies.

So, what should we invest in?

This begs the question: how and what do we invest in in turbulent markets?

Most people I know would probably point to yield farming on Berachain, Sonic, etc. — and that’s fine. But to me, there are many more interesting opportunities to explore, with a better risk-reward ratio, especially in times of crisis.

In my opinion, the most asymmetric bet right now is at the intersection of DeAI infrastructure and AI agents (more on that later).

Stick to the motto: “Be fearful when others are greedy, be greedy when others are fearful.” 

Crypto AI Sub-Sectors I’m Following

There are several particularly interesting sub-sectors in the crypto AI space right now in my opinion:

• Dev Tools – Frameworks, Vibe Coding Tools, MCP Infrastructure

• Decentralized AI Infrastructure – Decentralized Compute, Verifiability, Deployment, Confidentiality, Storage, Ownership

• Consumer AI – AI Agents, Alpha Tools, Gaming, DeFAI, GambleFAI, Personality/Companions (This is not all inclusive, but you get the idea.)

Framework Trends

For a more granular look at Consumer AI/AI Agents and Dev Tools, I created this post in March (originally planned to be monthly, but it seems the general agent market isn’t progressing fast enough to warrant a monthly update):

Despite this, we still see growth in open source frameworks and tools, such as @elizaOS (15,500 stars on GitHub), @arcdotfun (3.4k stars), and @sendaifun (1.2k stars), which received 434, 197, and 110 stars, respectively, last month.

Why Proxy Distribution Networks > Frameworks

I personally don’t find frameworks that exciting because they don’t have much value accrual. It’s much better to invest in distribution networks/proxy hubs because there is clear value accrual there - i.e. transaction fees from volume from speculators/investors trading AI proxy tokens. @virtuals_io is still the leader in this regard. Even with daily volume dropping from 8-9 figures to 7 figures, Virtuals is still the most trusted ecosystem for developers and the most diverse ecosystem with many teams trying to build unique proxy products.

@elizaOS is starting to look more interesting, especially now that @autodotfun (their launchpad) has just gone live. The team now has a distribution network that accrues value directly back into the $ai16z token.

What they need to solve is the execution problem of high-quality partner project launches to meaningfully differentiate from the service provided by Virtuals (otherwise they’ll be stuck with low-quality garbage projects with 4-5 figure market caps).

Anyway, taking a step back, while these AI agents, frameworks, and distribution networks are interesting, the area of ​​investment with the best risk-reward ratio right now is decentralized AI infrastructure.

Why?

If you’ve been in the AI ​​agent space for a while, you’ve probably noticed that agent products have progressed roughly like this:

Entertainment conversation “agent” ➔ Alpha analysis/tooling conversation “agent” ➔ Trading agent ➔ DeFAI abstraction layer ➔ Other smaller narratives ➔ Agents with smarter context, multi-agent/swarm, etc.

Death Flywheel Trap

The reason many teams get stuck is that there isn’t any proper core AI product in any of these “agent products”. The only AI is the chattering voice that automatically prompts the LLM every x amount of time.

Obviously, things have changed a lot from the early days, but the reliance on LLMs or off-the-shelf frameworks/workflows remains the same, so with every advancement/narrative in proxy products, secondary proxy products are created without proper use cases. (Similar to teams that forked major DeFi protocols a year later and faded away)

This has led to many teams creating hype with their proxy launches and their tokens, but then failing to maintain that attention (because there was no actual product), leading to a death flywheel (attention falls, token price falls).

Proxy builders need infrastructure; infrastructure builders need proxies

However, while these teams may fail, they are good at one thing - that is GTM (go-to-market strategy)/creating hype.

If there are many teams that are good at GTM of proxies, know how to play the token game/build a community, but lack proper AI products - what should they do? They should leverage professional AI models and machine learning capabilities from inference networks and DeAI infrastructure providers.

On the other hand, DeAI infrastructure teams are not good at GTM. They are not on the front lines, some are not crypto native, and don’t know how to build a community.

So… why not combine the two?

I believe the missing link between Deep AI infrastructure and viral agent distribution is where the real opportunity lies.

My Crypto AI Investment Thesis

This brings me to my Crypto AI Investment Thesis:

Invest in DeAI infrastructure and agent teams that introduce new, unique Web3 workflows that change the way people interact with existing crypto products (DeFi, on-chain).

In Web2, workflow automation and augmentation — increasing productivity (and thus profits) while minimizing costs — is very common in the agent vertical space, especially for mundane tasks (the more mundane, the higher the value). For example:

• Legal AI agents ingest raw paper documents, create legal case databases, and work with attorneys to help their clients succeed in court

• Accounting agents review receipts, invoices, general ledgers, trial balances, and generate unaudited financial statements and tax returns

• Architectural agents review building plans, estimate costs, and suggest ways to reduce construction/material costs while keeping durability and design consistent with client needs

There are many case studies like this in Web2 where startups quickly grew to 7-8 figure ARR (annual recurring revenue) in a few months because of this — they truly use AI agents to automate and enhance workflows and provide real value to other businesses/customers.

In Web3, this is still fairly new 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 it. The DeFAI abstraction layer solves this problem to some extent, but most are still unusable and have poor reasoning (you have to prompt very specific hints to make it work - which is actually counterproductive because ideally you want regular users to use it, and regular users usually don't know what they want to do, so naturally they don't know what to prompt).

This is why I think teams that can meaningfully change the Web3/crypto workflow are very rare. However, if you can spot them and invest in them early (now), you will get a lot of upside in the future.

On the other hand, we have DeAI infrastructure. Most of them are uninvestable because they are still in the early stages.

These teams tend to raise millions of dollars from VCs and take a few years to do a TGE (token generation event). Some of the projects that have launched have seen a 50-80% price drop due to market conditions. The ones that do well need to generate significant revenue to maintain the token price (or hire a very good market maker).

@getgrass_io is a great example - purportedly 8-9 figure revenues and a great consumer facing product (anyone can contribute bandwidth to get an airdrop).

Projects like Grass are very rare in VC backed DeAI infrastructure and usually the only way you can get involved early is by using the product/participating in the airdrop. They will likely drive up the token price at the TGE (low circulation, high FDV style) as VCs come in at relatively low valuations. If you decide to invest in similar projects, you are more likely to lose money than make money.

Investable Community First DeAI Ecosystem

This brings us to the alternative - a pure community/no VC DeAI ecosystem. Yep, that's Bittensor.

Before the dTAO upgrade, the ecosystem was pretty boring. Validators act as some kind of capital allocator as they decide which subnet gets $TAO emissions (capital).

But since the dTAO upgrade went live on Valentine’s Day this year, that dynamic has changed dramatically. Now the market decides which subnets get emission. The community — the people — are now the capital allocators. If the community believes your subnet has no product and doesn’t provide much value, you don’t get emission (capital). This drives subnets to build publicly, launch faster, and build products that people actually want.

@BarrySilbert is betting on the Bittensor ecosystem through @YumaGroup (a subsidiary of DCG), which invests in, builds, and incubates Bittensor subnets. A recent interview between @RaoulGMI and @BarrySilbert has created a lot of excitement in the community (because a major crypto institution has now entered the Bittensor ecosystem):

From an investment perspective, the liquidity of the Bittensor ecosystem is much better than that of the AI ​​proxy ecosystem. The core problem with proxy ecosystems like Virtuals is that LPs are paired with Virtuals, which results in higher volatility and more impermanent loss for liquidity providers.

This is why liquidity is often low — you can typically only deploy $1k-5k and experience 3-7% slippage on these proxy tokens. On the other hand, deploying a similar amount into a subnet token would only incur 0.05%-0.1% (or even less) slippage.

Quick Summary:

• Crypto AI agent hype cycle is fading, true products + user retention are still rare

• DeAI infrastructure is underestimated, misunderstood and mispriced

• The best strategy is to combine infrastructure + agent GTM to unlock new workflows

• $VIRTUAL leads the agent metaverse, Bittensor leads the infrastructure metaverse

• Watch for teams merging the two - huge upside if found early

Summary

I believe DeAI will define the next trend in Web3 AI. We will see more teams changing the way we interact with each other and protocols, changing the way value is created, and generating more new areas that reach more users and occupy more market share (more mainstream). Now is the time to quickly understand DeAI infrastructure and how it changes things. Be sure to keep an eye on teams that can successfully combine DeAI and proxies.

Remember, my theory is not set in stone. I am constantly learning and refining it. I do my best to ensure we can catch the next big trend in Web3 AI. Again, this is not financial advice - please do your own research and take everything in this article with a grain of salt.