Recently scrolling through the timeline, I see a bunch of people shouting about DeFAI.

At first, I also thought it was just an upgraded version of “AI helps you play DeFi.”
But after doing my own research, it feels like it’s not that simple, and it’s not that fast either.
Let me start with my own intuition:
Right now, most of the so-called DeFAI
In essence, it’s still “automation tools + wrapping a layer of AI skin.”
Still a long way off from a true “Agent helping you manage your money.”
This narrative—how it really took off—
Actually follows a pretty typical three-step process:
First, an AI bot goes viral (the kind that can post on X and make memes).
The market’s first realization is: AI can also participate in the on-chain economy.
Then it starts to assetize it—
Issuing the Agent’s own tokens, doing valuation, and building ecosystems.
Only then do they start talking about execution,
Meaning letting AI actually move your money on-chain.
The problem is—
The first two steps moved too quickly.
The third step is just beginning.
When I look at these projects, I focus on one thing:
Does it really “touch money”?
If it’s only helping you analyze and give signals,
Then it’s fundamentally no different from the previous quantitative tools.
Once it starts executing trades,
The risk model is a completely different level.
The current technical path is also pretty realistic:
AI calculates off-chain,
and on-chain is just for execution.
To put it plainly,
Agents aren’t as “autonomous” as you think.
It’s more like a:
Thinking trading script.
But there’s a key point that many people overlook:
As long as it can move your assets,
it will definitely involve permissions, private keys, and signatures.
If anything goes wrong here,
then all the strategies above are 0.
So my current judgment on this track is:
Technology isn’t the bottleneck.
Trust is.
From what I can see at the application layer, it basically boils down to four things:
1） Automatically finding yield (helping you do arbitrage across different protocols)
2） Quantitative strategies with automated execution
3） Making trades using chat commands
4） Helping you watch your positions to prevent liquidation
The first two already have people running them.
The last two are still in the early stages.
But honestly—
the one with real value is the most boring one:
Risk control + monitoring.
Because it doesn’t directly move money,
it has a higher tolerance for errors,
and it’s also easier for institutions to accept.
There’s another point I’ve been thinking about recently:
Many people imagine AI helping you “make money automatically,”
but the reality is:
Even in traditional finance,
Agents haven’t been run through at scale yet.
Not to mention on-chain, with such a high-volatility + high-risk environment.
So at this stage, my attitude is very simple:
DeFAI = full of imagination, but slightly overhyped.
In the short term, focus on the narrative.
In the medium term, look at who can truly produce stable strategies.
In the long term, look at who can solve the “trust in execution” issue.
If I absolutely have to bet on one direction:
I’m more inclined toward
starting with “assistants” rather than “replacements.”
For example, watching positions, pre-emptive alerts, and governance analysis.
Then gradually transitioning to execution.
This track will definitely produce something,
but it probably won’t happen overnight.
Right now, it’s more like—
everyone is talking about the future,
but no one has really handed their money to AI for a long time.
I also still haven’t dared to put all my funds in.