An AI agent can research a market, analyze data or call on another piece of software to finish a task. But when those services cost money, who handles the checkout?
Sui and Alibaba Cloud are working on a system that lets the agent pay within a budget its user sets.
Announced on October 7 at Sui Basecamp in Singapore, the collaboration would bring Alibaba Cloud services into Sui Agent Payments, allowing agents to buy cloud and AI services with payments settled on Sui. The announcement came during TOKEN2049 week, as companies gathered in Singapore to show how digital assets could support everyday business activity.
The idea is straightforward: give an agent a task, define what it can spend and let it purchase the resources needed to complete the job.
Post from @SuiNetworkView the original post on X↗
A Budget For Getting The Job Done
Imagine asking an agent to prepare a market report. It might need to buy a dataset, run calculations and use an AI model to turn the results into something readable.
Under this model, the agent could pay for approved services as it uses them, drawing from an assigned budget. Sui describes the planned Alibaba integration as supporting payments in stablecoins for individual service requests. Alibaba says access would operate within defined controls.
For a business, the potential benefit is less manual work between deciding what needs doing and obtaining the resources to do it.
The important change is who can authorize and complete each purchase. An employee would establish the boundaries, while software handles spending within them. That could become useful when an agent needs several services to complete one assignment.
Whether it saves money will depend on pricing and how efficiently the agent uses those resources.
Who Gave The Agent Permission?
A spending limit answers one question: how much can the agent spend?
It leaves others open. Which services can it buy? Who can receive the money? When does its permission expire?
Sui’s broader approach to agentic commerce includes authority limited by amount, timing and recipient. Those controls help distinguish an agent carrying out a specific assignment from software with unrestricted access to funds. The precise configuration for Alibaba Cloud’s integration remains to be detailed.
This connects with the argument Chandler Fang makes in today’s Blockster opinion piece on the XRP Ledger and the agentic economy: faster payments need identity checks, spending controls and a way to establish whether an agent is following its user’s instructions.
An agent could remain within its budget and still buy the wrong thing. Businesses need controls around the purpose of a payment as well as its size.
Google Cloud Enters The Accountability Side
A separate announcement this week addresses what happens when someone needs to check an agent’s work.
On October 6, Mysten Labs announced a collaboration with Google Cloud on Verifiable Agent Arbiter, or VAA. The system is being developed to provide independently verifiable records of what an agent was allowed to do, what it did and what resulted.
Detailed records would remain in customer-controlled Google Cloud Storage. Cryptographic proofs—used to check that those records have not been altered—would be stored on Walrus and coordinated through Sui.
That could help companies investigate an incorrect purchase, a disputed service or an agent acting outside its instructions. Mysten says initial enterprise deployments are planned before broader availability.
Post from @Mysten_LabsView the original post on X↗
Google’s Earlier Moves Help Explain The Direction
Google’s involvement follows a pattern Blockster has been tracking.
In September, we covered its recruitment for stablecoin and tokenization expertise, including knowledge of digital payments, tokenized deposits and custody.
We then reported on Google Cloud’s work with Puffer. That partnership focuses on giving applications early assurances about transaction execution before final settlement on Ethereum.
The Mysten collaboration adds another piece: evidence that helps businesses check how agents used their authority.
These are separate initiatives, but together they suggest Google is building both the expertise and infrastructure needed for software to handle financial activity. Alibaba’s collaboration brings another major cloud provider into that picture, with a focus on how agents purchase computing services.
The Next Step Is Putting It To Work
For Sui and Alibaba Cloud, the practical questions now concern availability, supported services and pricing. Businesses will also want to know how they can stop an agent’s spending and what happens when a paid request fails.
The opportunity is easy to understand. Software that can obtain the resources it needs could complete more work without waiting for a person at every checkout.
Making that useful requires clear budgets, reliable services and records people can inspect when something goes wrong.
The agent may handle the checkout. The person funding it still needs to stay in control.
Reporting by Lidia Yadlos
