AI Is Moving From “Assistant” to “Actor”

AI has been primarily used for the past several years to get things done for us.

Write this.

Summarize that.

Analyze this chart.

Let's find the lowest price.

The following phase is quite different, however.

Rather than providing answers, AI agents are beginning to act.

It is possible for an agent to search for a service, compare the various options, communicate with another system, and complete a transaction within the scope it has been given. Agentic commerce is emerging from the realm of the future towards actual infrastructure, and the companies that are pursuing the concept are increasingly paying attention to identity, controls and authorizations for payments driven by AI.

That sounds exciting.

Until there is trouble.

Suppose you give an AI agent the ability to handle a budget.Let's say you hand over a budget to an AI agent.

Perhaps she can spend $500.

It finds a service.

The other AI agent negotiates the deal.

A payment is activated.

The deal is done.

Then you find that the service is false, the agent has misunderstood your instructions or the transaction should never have taken place.

Now the awkward query:

Who do you blame?

The user?

The AI developer?

Who is the company that operates the agent?

The wallet provider?

The payment network?

Or the AI itself?

One of the largest challenges of the agent economy could be that question.

The Missing Layer: Accountability

The Web was designed for humans and organizations.

A website can identify the person who is logging onto it.

A bank knows the account owner.

A business may enter into a contract.

With AI agents, however, there is a weird new dynamic.

They are no longer just companies or traditional legal entities, but increasingly can carry out actions with real economic implications without being human.

That's what I believe to be the missing ingredient in the agent economy.

Not intelligence.

Not faster models.

Not even payments.

Accountability.

The problem is being identified more and more: just knowing what an AI agent did is no guarantee. For complex systems, it's also important to set up who authorized the action, if the authorization was valid throughout a multi-agent workflow and who is responsible for the final responsibility.

It's here that the discussion gets interesting.

For the reason that crypto may need to be a part of the answer.

Crypto Might Finally Have a Job Bigger Than Speculation

The crypto industry has been known to capitalize on every trend for years.

AI became popular?

All of a sudden, hundreds of projects wanted to be "AI tokens.All of a sudden hundreds of projects wanted to be "AI tokens".

That was mostly a tale of empty words.

However, there could be a more practical relationship between AI and blockchain.

AI agents must be able to establish things.

Who is this agent?

Who authorized it?

What restrictions does it have?

What did it do?

When did it happen?

Did this record change later?

Blockchain is not magic and moving AI “on-chain” is not the answer to these issues.

It can potentially offer some of the infrastructure: programmable wallets, transaction records, cryptographic verification, and auditable trails: those are all things that a blockchain-based system can offer. The research agenda for agent-to-agent finance is investigating these problems (the ability of autonomous agents to interact economically, while maintaining identifiable, authorized and accountable agents).

This is a much more interesting use case than just launching another token with the word “AI”.

AI Agents Need Wallets — But Not Unlimited Power

That's where things might get dangerous.

It's easy to see how an AI agent can have access to money.

It's a bad idea to allow it to have unlimited access.

The future is likely to be:

Here is my wallet; good luck!

Rather, it could be something more akin to this:

An agent is assigned a particular identity.
It is placed in a separate wallet.
It has a budget cap.
It can only communicate with authorized services.
Increased approval is needed for larger transactions.
All key actions are documented.

The concept of bounded autonomy is gaining traction in the discourse on trustworthy AI systems. The idea is not that it will prevent agents from doing what they do, but that it will make it clear what they can and cannot do before doing it.

It's like giving an employee a company card.

You don't permit them to use the complete bank account and tell them:

“Do what you feel is doing the best.”

You set boundaries for them.

Rules.

Permissions.

Oversight.

Exactly that is what AI agents might require, but at machine pace.

The Real AI Arms Race Might Be About Trust

All are racing to create more intelligent artificial intelligence.

Bigger models.

Better reasoning.

More autonomous agents.

However, at some point intelligence may not be the major issue.

Trust will be the limiting factor.

Let's pretend that there are two AI agents.

Agent A is a very smart agent.

Agent B has some intelligence but he has a known name, some permissions, a transparent transaction history and some spending limits.

Which would you give your hard-earned cash to?

Most people know the answer, and it's the right one.

Even the most intelligent AI around the world is hard to trust without an explanation:

By whose authority?

What could it do?

What actually happened?

But who's on the hook if it doesn't?

As increasingly meaningful tasks become the responsibility of autonomous agents, industry and policy debate takes an increasing turn toward questions of authorization, identity and meaningful human control.

The Next Crypto Boom Might Not Look Like the Last One

The next big crypto story may not start out with a meme.

It may start with a quiet rumble.

With infrastructure.

Agent identities.

Programmable wallets.

Machine-to-machine payments.

Verification systems.

Reputation.

Authorization.

Dispute resolution.

Audit trails.

Not all AI and crypto-related projects will be successful, as is not the case with that one.

Far from it.

Indeed, this may generate another round of buzz, scams and worthless tokens that claim to fix issues they don't really comprehend.

But there is a real question that is coming up under the cacophony.

What will enable humans, companies and other machines to trust AI agents, if they are going to be economic actors?

That is where blockchain might actually be more use than speculative.

Not because blockchains make AI smarter.

However, an increasingly independent AI economy could require systems that can keep track of permissions, confirm actions, and provide evidence if something goes wrong.

The Biggest Question Isn't “Can AI Do It?”

We are getting closer and closer to a time when AI can do things that we used to think were only humans' privilege.

Search.

Analyze.

Negotiate.

Choose.

Pay.

Execute.

Technological question is in a process of transformation.

It's not about "it" anymore:

“Is it possible for AI agents to do this?”

The bigger question might be:

“Should this AI agent be allowed to do this—and how can we prove it had permission?”

It could be the missing piece of the whole agent economy.

And the companies that get it solved may be much more significant than the companies just creating yet another chatbot.

AI is becoming more knowledgable for action.

Crypto is becoming proficient in the skills of verification.

Possibly there is a large next chance where the two worlds merge.

Not in another AI Token.

However, in the infrastructure that answers one simple question:

Who is Really in Control When an AI Agent Acts?

Final Thought

The AI revolution may not be limited by intelligence.

It could be a lack of trust that is the problem.

And when trillions, eventually billions, of dollars begin to flow through independent software, the most valuable technology could be the one that's making the decisions.

It might be the infrastructure that will determine who approved it, what it allowed to do, and who will be responsible after.

Smarter AI is not enough for the agent economy.

It must have a means of believing it.