The rapid rise of agentic AI is creating a new question for the blockchain industry: can autonomous software become an active participant in the economy, rather than simply a tool for generating information?
Franklin Templeton Digital Assets estimates agentic commerce could reach $3 trillion to $5 trillion by 2030, but significant questions remain around payments, security, accountability and trust.
Aptos Co-Founder and CEO Avery Ching argues that the infrastructure supporting AI agents will need to handle transactions that are fast, low-cost, programmable and auditable.
In this interview with Invezz, Ching discusses what that infrastructure could look like and where blockchain fits into the emerging agent economy.
Invezz: Franklin Templeton estimates agentic commerce could reach $3–5 trillion by 2030. What has to happen over the next few years for that forecast to become reality?
I think the bigger question is whether agents become reliable enough that people and businesses are actually comfortable giving them authority.
Models are powerful enough for most of our everyday workflows. The remaining problems are around the infrastructure.
Can an agent keep your data confidential, can you deploy them easily, can they run reliably for long periods of time, can you prove what they did? And can they hold and exchange value under clear rules?
For commerce specifically, agents need their own identity and payment capabilities, but they also need boundaries.
You should be able to give an agent access to capital with a spending limit, define who it can transact with and what it can buy, and have a provable record of what happened and why.
Once you have that, the agent can move from recommending a transaction to actually completing one, and at that point the economics matter too.
If agents are making huge numbers of very small payments, those transactions need to be fast and cheap enough that the payment itself doesn’t become the bottleneck.
Invezz: Aptos tops Franklin Templeton’s throughput rankings, but is TPS really the deciding factor for AI commerce, or are latency, reliability, and costs more important?
Transaction throughput is one part of it, but we have to consider the entire system properties.
What matters is whether the overall decentralized financial system is secure, fast, scales, and low cost. Aptos is the best technology to meet this high bar.
Agents are an extension of our personal and corporate lives.
The infrastructure needs to look much more like mature cloud and database infrastructure: durable state, recovery, failover and predictable performance.
Aptos was designed around that kind of high-throughput, low-latency environment, which is why I think it is ideal for machine-driven activity.
Invezz: What do you see as the first real-world use case where AI agents will be making payments autonomously at scale?
I think it starts with agents transacting with money.
An agent paying for goods and services using the same interfaces we do today is the first step. Today, agents are mostly read-only in an economic sense.
They can find information and recommend an action, but another system or a person usually has to complete the transaction.
Once an agent can pay another service directly on a per-request basis, you get a different kind of network.
The agent can discover what it needs, pay for it, and continue the task without setting up an account, negotiating a contract, or waiting for someone to approve every individual payment.
After this grows, agents will start to negotiate and transact with each other at machine speeds. That’s where I think machine-to-machine commerce starts to become real.
Invezz: If an AI agent makes a bad purchase or is compromised, who should ultimately be accountable—the user, the developer, or the underlying infrastructure?
There probably isn’t one answer for every situation, but the system should make it possible to know exactly what happened.
That’s the first requirement. If an agent spends money, accesses data, or sends information somewhere, you should be able to trace the action back to the policy and authority it was operating under, with an immutable ledger.
The user or enterprise should define what the agent is allowed to do. The developer needs to enforce those controls correctly.
And the infrastructure should give you a record that neither side can quietly rewrite later.
On the payment side, that means an agent shouldn’t just have unrestricted access to a wallet.
It can have its own keys, spending caps, approved counterparties, or categories of activity.
The interesting part of programmable money is that you can give an agent enough authority to be useful without giving it unlimited authority.
As agents become more autonomous, I think that auditability becomes just as important as the payment itself.
If you can’t prove what an agent did and why it was allowed to do it, enterprises are going to be very reluctant to give it meaningful control.
Invezz: Large AI companies already rely on centralized cloud infrastructure. Why do you believe they’ll adopt blockchain-based payments instead of simply improving existing payment systems?
I don’t think every AI interaction needs a blockchain, and I don’t think that’s the right way to frame it.
The question is what happens when software needs to transact with software it doesn’t already have a commercial relationship with.
Existing payment systems work very well for a person or business with an account, an identity, a payment processor, and an established counterparty.
Agents create a different pattern. They may need to discover a service, pay a fraction of a cent or a few cents for it, receive the result and move on, potentially millions of times across many different counterparties.
Setting up accounts, API keys, invoices, and bilateral relationships for every interaction doesn’t scale particularly well.
Open payment rails become useful because an agent can hold value and transact programmatically under a defined policy.
Stablecoins give you a digital unit of account, and protocols like x402 make per-request payments possible.
The blockchain is useful here because it gives unrelated machines a common settlement and accountability layer without requiring one company to operate the entire network.
Invezz: Blockchain has seen several narratives over the years, from DeFi to NFTs and gaming, that took longer than many expected to reach mainstream adoption. What makes agentic AI different, and what evidence are you seeing today that gives you confidence?
I would still be careful about assuming a timeline. We’re very early. What is different is that the demand for agents exists independently of crypto.
People are already using increasingly capable AI systems, enterprises are deploying them, and model quality is improving very quickly.
The adoption of AI will be exponential when we solve the challenges of confidentiality, scaling, audit trails, and ease of use.
The harder problem is turning an agent from something a relatively technical person can experiment with into something an ordinary person or enterprise can actually trust and depend on.
It needs to be confidential, easy to use, reliable, accountable, and capable of paying for things.
Those pieces all exist in some form today, but they haven’t really been assembled into one system yet.
That’s what gives me confidence in the opportunity, but it’s also why I wouldn’t say the market is already here.
We are still building the infrastructure that lets agents move from demos and copilots into software that can operate continuously and take meaningful actions on someone’s behalf.
Invezz: This vision relies heavily on stablecoins for machine-to-machine payments. How resilient is that model if a stablecoin experiences volatility or faces regulatory disruption, and what safeguards should exist to ensure transactions can continue safely?
I wouldn’t design the system around one stablecoin or assume one asset works everywhere.
The important thing is that the agent has a clear policy around what it is allowed to hold, what it is allowed to spend, and under what conditions.
You can imagine an agent wallet with approved assets, spending limits, approved counterparties, and rules that automatically stop activity if an asset or counterparty falls outside the policy.
That becomes especially important because the goal isn’t to have a human approve every transaction.
The safeguards have to exist in the system before the agent acts.
Stablecoins are useful because they give software access to digital money that can move globally and programmatically, but they’re one piece of the architecture.
The more important principle is that agents need a secure way to hold and exchange value without giving them unlimited financial authority.
If we get that right, the payment rail underneath can evolve over time.
The post Interview: Aptos CEO on AI agents, stablecoins and the future of machine commerce appeared first on Invezz
