AMCapitalX

The more I watch AI and crypto move closer together, the more I think we’re focusing on the wrong thing.
Most of the conversation is about how capable AI agents are becoming. They can monitor markets, trade tokens, move money, search for yield, rebalance portfolios, and make decisions faster than any person could. That is impressive. I don’t want to take anything away from that.
But honestly, capability isn’t the part that worries me.
Trust is.
The moment an AI agent gets access to a wallet, the whole conversation changes. It’s no longer just answering questions or suggesting what someone should do. It can actually act. It can move funds. It can sign transactions. It can interact with protocols. It can make a decision that has a real financial consequence.
And in crypto, that consequence can be permanent.
That is why I think the industry needs to slow down, at least mentally, and ask a more uncomfortable question: just because an agent can do something, should it be allowed to?
For me, that is where Newton’s idea around guardrails starts to make sense.
I’m not interested in guardrails because I think AI agents should be weak or heavily controlled. Actually, I think agents will become much more useful as they gain more independence. But independence without limits is not something I would call progress. In finance, I would call it risk.
Real professional environments already understand this.
A trader can have freedom, but there’s usually a mandate. A finance manager can approve payments, but there are limits. Someone running a treasury may be trusted with serious responsibility, but that doesn’t mean they can send every dollar to any account they choose.
That isn’t a lack of trust.
That is how trust works in practice.
You give someone enough authority to do the job, but you also define where that authority ends.
I don’t see why AI agents should be treated differently.
In fact, I think the need for limits is even stronger with AI.
Human beings don’t always give clear instructions. We say things like, “Find a better return, but don’t take too much risk.” A person with experience will understand that this sentence is incomplete. They’ll probably ask questions.
How much risk is acceptable?
Can we use leverage?
Can money be moved to another chain?
Can the strategy use a new protocol?
How long can the capital be locked?
What happens if the market becomes unstable?
These are normal questions.
But an AI agent may treat the instruction differently. It may simply try to solve the task as efficiently as possible.
That’s where the danger starts.
The user has an intention. The machine has an instruction. Those two things can look similar, but they’re not always the same.
I’ve noticed this in the broader AI conversation too. People often assume that if a system is smart enough, it’ll somehow understand what we really meant.
I’m not sure that’s a safe assumption, especially when money is involved.
A poorly understood email can be corrected.
A bad transaction may not be.
That’s why I like the idea of separating the agent’s decision from the final authorization.
Let the agent think. Let it search widely. Let it compare opportunities. Let it react quickly.
But before money actually moves, there should be a clear check: is this action within the rules?
That could mean a spending limit.
It could mean only using approved protocols.
It could mean blocking transfers to unknown addresses.
It could mean restricting how much capital can be put into one asset.
It could mean requiring extra approval above a certain amount.
To me, that doesn’t make the agent less useful.
It makes the agent easier to trust.
And I think trust is what will matter when AI agents move beyond experiments and start handling serious capital.
Right now, a lot of agent demos are impressive because they show action. An agent spots an opportunity, makes a trade, moves across protocols, or adjusts a position.
But in a real business, people won’t only ask whether the agent can act.
They’ll ask what happens when it makes a mistake.
That question is much harder to answer.
What happens if the agent receives bad data?
What if a contract behaves in a way it didn’t expect?
What if someone tricks the system with a malicious instruction?
What if market conditions change quickly?
What if the agent follows the words of the instruction but completely misses the user’s intention?
And the biggest question of all: who is responsible when the money is gone?
Those questions are not anti-innovation.
They are the questions that show up when a technology starts becoming real.
I’ve seen this pattern many times. At the beginning of a new technology cycle, people care about freedom and possibilities. Rules feel boring. Safety feels like something that can be solved later.
Then the technology gets bigger.
More people start using it.
More money gets involved.
And suddenly the boring questions become the important ones.
Who has access?
Who is responsible?
What are the limits?
Can the system be stopped?
What happens when something fails?
Crypto itself has gone through this cycle more than once. After every major failure, the industry returns to custody, permissions, security, governance, audits, and risk management.
AI agents won’t somehow escape those issues.
They may actually make them more difficult.
One reason is speed.
Speed is one of the biggest advantages of an AI agent. It can act faster than a person. It can watch the market while people sleep. It can respond to changes immediately.
But speed works both ways.
An agent can make a good decision quickly.
It can also make a bad decision quickly.
And worse, it can keep acting before anyone notices there is a problem.
A person might make one bad trade and then stop to think.
An automated agent could make several connected decisions, move funds, enter positions, and interact with multiple protocols in the time it takes a human to understand what happened.
That’s why I don’t think the answer is simply keeping a person in the loop for every transaction.
That doesn’t really work either.
Imagine having to manually approve every small payment or every portfolio adjustment an agent wants to make. At that point, you lose much of the value of having an agent.
The better approach, in my opinion, is to approve the boundaries instead of approving every action.
That’s already how most organizations work.
A manager gets a budget.
A trader gets a mandate.
A team gets a set of permissions.
People don’t go back to the CEO every time they need to make a normal decision.
The rules are already there.
AI agents should probably work in the same way.
Give them space to act, but make the boundaries clear.
That middle ground feels far more realistic than the two extremes we often hear about.
One extreme is total human control, where an AI agent can barely do anything without asking.
The other is total machine freedom, where the agent has access to funds and almost no meaningful limits.
I don’t think either one is practical.
The future is probably controlled delegation.
To me, that means an owner, company, fund, or institution decides what the agent is allowed to do. The agent can then act independently inside those limits.
That is the model I would be more comfortable with.
Still, I don’t think guardrails are a perfect solution.
Guardrails can fail too.
A badly written rule can create problems. A policy can be too strict and block useful actions. It can be too loose and allow dangerous ones. The system enforcing the rules can have bugs.
And then there is the question of control.
Who decides the rules?
Who can update them?
Who controls the data used to make decisions?
Can a company change the system in a way that users don’t expect?
These questions matter, especially in crypto.
The whole industry was built around reducing unnecessary trust in middlemen. So it would be strange if the future of AI agents depended on one central company deciding what every agent is allowed to do.
That’s not the kind of guardrail system I would want.
I think the better model is one where users define their own boundaries and the infrastructure simply enforces them.
There’s a real difference between asking someone else for permission and creating your own mandate.
A company should be able to say: this agent can move this amount of money, use these protocols, deal with these counterparties, and stop under these conditions.
Then the system should enforce that.
The rules should belong to the owner of the capital.
That, to me, is what makes the idea interesting.
It is not about stopping AI agents.
It is about making delegation more precise.
And I think precise delegation is going to matter a lot more than people realize.
The AI + crypto conversation often makes everything sound futuristic, but the underlying problem is very old.
How do you give someone power without giving away all control?
Companies have been dealing with that question for centuries.
Banks deal with it.
Investment firms deal with it.
Governments deal with it.
Families deal with it.
Any time one person gives another person authority over money, limits appear.
The technology may be new, but the problem isn’t.
What changes with AI is the speed, the scale, and the fact that the agent may behave in ways we didn’t fully predict.
That is why I believe authorization will become one of the most important parts of the AI and crypto stack.
We’ll still need smarter agents.
We’ll still need better models.
We’ll still need faster infrastructure and better user experiences.
But none of that will matter for serious adoption if people are afraid to let the agent act.
Trust is the real bottleneck.
And trust doesn’t mean believing that the AI will never make a mistake.
That isn’t realistic.
To me, trust means knowing that one mistake cannot become an unlimited disaster.
That is a very different idea.
I’m positive about the future of AI agents.
I can imagine them handling routine treasury work, monitoring positions, managing payments, searching for better capital efficiency, and helping smaller teams do things that once required large financial departments.
I think that future is coming.
But I don’t think it will arrive through blind confidence in AI.
It will arrive because the systems around the AI become better.
Clearer permissions.
Better checks.
Better limits.
More transparency.
Better ways to understand who authorized what.
That is why Newton’s direction interests me.
I’m not saying Newton will definitely win this market. It’s far too early to say that. There will probably be different approaches, different systems, and different standards.
Some will focus on DeFi.
Some will focus on payments.
Some will be designed for institutions.
Some may be open and decentralized.
The market will decide what works.
But I do believe the problem Newton is trying to address is real.
AI agents need more than intelligence.
They need boundaries.
The mistake we should avoid is confusing a machine’s ability with its authority.
An agent might be smart enough to identify an opportunity.
That doesn’t mean the opportunity fits the user’s risk tolerance.
It might be capable of sending money.
That doesn’t mean it should be able to send any amount to anyone.
It might act faster than a human.
That doesn’t mean faster is always better.
These distinctions may sound obvious, but I think they will define the next stage of AI-powered finance.
The smartest agent may find the opportunity.
The fastest agent may get there first.
But the agent trusted with serious money will be the one that can show where its freedom begins and where it ends.
That’s why I don’t see guardrails as something holding AI agents back.
I see them as the point where AI agents become useful enough, safe enough, and mature enough to be trusted in the real world.
@NewtonProtocol $NEWT #Newt

リアルワールド・アセット（Real World Assets、RWA）について考えるとき、私は技術からは始めません。
私は人から始めます。
私は、ある資産（プロパティ）、債券、またはローンがトークン化され、すぐに好奇心をかき立てられる相手のことを考えます。ダッシュボードで高いリターンを見て、それが本当に価値ある機会なのか、それとも新しいテクノロジーで装った単なる複雑な金融商品なのかと疑う人のことを考えます。スマートコントラクトは理解しているのに、不動産法、破産、信用リスクについて真剣に考えたことのない開発者のことも考えます。そして、それらすべてを理解しているのに、ウォレット、ブロックチェーン、分散型システムを信じるのが難しいと感じている伝統的な金融の専門家のことも考えます。