As AI agents begin searching, negotiating and purchasing on behalf of people and enterprises, the strategic question is no longer whether machines will transact—but how human authority, payment execution and accountability will be connected.
Artificial intelligence is rapidly moving beyond answering questions and generating content. The next phase of AI will involve autonomous agents that search for products, negotiate with service providers, book appointments, purchase software, procure business supplies and execute financial transactions on behalf of individuals and organisations.
Circle CEO Jeremy Allaire has predicted that, within the next three to five years, there could be billions of AI agents continuously conducting economic activity worldwide. This raises a foundational question: What payment infrastructure will these agents use?
Stablecoins are frequently presented as the natural answer because they are programmable, digitally native, globally transferable and capable of supporting small machine-to-machine transactions. However, stablecoins are unlikely to be the only payment rail for the emerging agentic economy.
In India, the Unified Payments Interface, or UPI, has the potential to become the most natural domestic payment interface for AI agents purchasing goods and services.
UPI already provides many of the characteristics required for agentic commerce: real-time settlement, interoperability, widespread merchant acceptance, low transaction costs, regulated bank-account connectivity and an API-driven ecosystem.
The next step is to transform UPI from a payment system initiated by humans into a programmable transaction network capable of executing payments under precisely defined human authority.
The payment would no longer represent an isolated instruction issued at the moment of purchase. It would represent the execution of a previously authorised economic objective.
From Human Payments to Agentic Commerce
Today, most digital payments begin with a human action. A person opens a mobile application, scans a QR code, enters an amount and authenticates the transaction.
Agentic commerce will work differently. A user may instruct an AI agent: “Purchase my regular groceries every Saturday, keep the total below ₹2,500 and use only approved merchants.” The agent could then compare prices, check inventory, select products, arrange delivery and complete the payment without requiring the user to manually participate in every step.
At the enterprise level, an authorised procurement agent could purchase cloud services, software licences, data subscriptions or office supplies within a predefined budget and approved vendor framework. This would fundamentally change payments.
Why UPI Is Well Positioned
UPI is already one of the world’s most advanced real-time retail payment systems. It connects regulated bank accounts through an interoperable network used by consumers, merchants, banks and payment applications.
Its advantages for agentic commerce include:
- Instant bank-to-bank payments.
- Broad consumer and merchant acceptance.
- API-based interoperability.
- Low-cost transaction processing.
- Strong linkage to verified users and regulated financial institutions.
- Support for mandates and recurring payments.
- An established framework for dispute resolution and transaction monitoring.
UPI has also started moving beyond payments initiated exclusively through smartphones. UPI Circle introduces the concept of delegated payment authority, allowing a primary user to authorise another user to make transactions within predefined controls. This model is highly relevant to agentic payments because an AI agent is, in effect, a delegated economic actor. UPI Reserve Pay provides another important building block. It enables a user to reserve or authorise an amount in advance, allowing subsequent payments to be processed as goods or services are delivered without requiring repeated authentication for every transaction. The combination of delegated authority, programmable mandates and real-time settlement creates the foundation for a future Agentic UPI ecosystem.
UPI Circle Is a Beginning, Not the Complete Solution
UPI Circle was primarily designed for one human user to delegate payment authority to another human user. Delegating authority to an AI agent is considerably more complex.
An AI agent should never receive unrestricted access to a person’s bank account. It should receive a limited, purpose-specific and revocable mandate. For example: “My grocery agent may spend up to ₹8,000 per month at approved grocery merchants, with a maximum of ₹2,000 per order. It may not purchase alcohol, gift cards or subscription services. Purchases from new merchants require my approval.” Such an authorisation would need to be machine-readable and cryptographically connected to several controls:
- The identity of the person granting authority.
- The identity of the authorised AI agent.
- The specific version or configuration of that agent.
- Approved merchants and merchant categories.
- Maximum transaction and cumulative spending limits.
- Permitted goods and services.
- Time, location and frequency restrictions.
- Human-approval thresholds.
- Refund and cancellation conditions.
- Expiry and immediate revocation mechanisms.
The user’s UPI PIN should never be shared with, stored by or reproduced by an AI agent. The PIN should only be used by the human account holder to establish, modify or approve the underlying mandate. Once the mandate has been created, the agent should operate only within its authorised boundaries.
What Agentic UPI Could Look Like
A future Agentic UPI transaction could follow this sequence:
HUMAN INTENT → DELEGATED AUTHORITY → AGENT ACTION → POLICY EVALUATION → PAYMENT EXECUTION → SIGNED EVIDENCE → SETTLEMENT & MONITORING
Consider a consumer asking an AI agent to order household groceries. The user first creates a mandate defining the monthly budget, maximum order value, permitted merchants and product restrictions. The agent then searches for products, compares prices and prepares a shopping cart. Before payment is initiated, an independent authorisation gateway evaluates the proposed transaction against the user’s mandate. Transactions that comply with the authorised policy are approved automatically. Transactions involving a new merchant, restricted item, unusual amount or suspicious pattern are escalated to the user for human approval. UPI then completes the bank-to-bank payment.
Finally, a tamper-evident record captures the complete transaction chain, including the user’s authority, the agent’s identity, the selected products, the applicable policy, the authorisation decision, the payment result and the merchant’s fulfilment outcome. This is significantly more sophisticated than simply connecting an AI chatbot to a payment API.
The Real Challenge Is Authorisation, Not Payment
The payment rail is only the final execution layer. The more difficult problem is proving that an AI agent had legitimate authority to conduct the transaction.
Every agentic payment system must answer several critical questions:
- Who authorised the agent?
- What exactly was the agent permitted to purchase?
- Did the transaction remain within the delegated mandate?
- Was the merchant verified?
- Was the agent or its execution environment compromised?
- Did the agent manipulate or misunderstand the user’s intent?
- Who is liable if an incorrect or unauthorised purchase occurs?
- Can the transaction be revoked, disputed or reversed?
- Can an auditor independently verify the complete decision chain?
Traditional payment authentication proves that a payment credential was used. Agentic commerce requires something more: proof that the transaction was consistent with the user’s intention, organisational policy and delegated authority. This will require deterministic policy enforcement, cryptographic identity, transaction-level risk assessment, continuous authorisation and verifiable evidence.
UPI and Stablecoins Will Serve Different Roles
The future of agentic payments is unlikely to be controlled by one universal payment system. UPI and stablecoins offer different advantages.
UPI is particularly well positioned for regulated, rupee-denominated domestic payments involving Indian consumers, merchants and enterprises. It provides direct integration with regulated bank accounts, familiar consumer protections and widespread domestic acceptance.
Stablecoins may be more suitable for international machine-to-machine payments, cross-border transactions, programmable digital assets and extremely small payments for services such as computing capacity, API access, data and digital content.
An AI agent operating in India could use UPI to purchase groceries from a domestic merchant while using a stablecoin to pay a foreign data provider or another software agent. The likely future is therefore multi-rail.
Payment orchestration platforms will dynamically select between UPI, bank transfers, cards, central bank digital currencies and stablecoins depending on the transaction’s currency, jurisdiction, value, merchant, risk and regulatory requirements.
In this emerging architecture:
- UPI could become the native rail for regulated domestic agentic commerce.
- Stablecoins could become an important rail for global machine commerce.
- Authorisation and proof platforms would govern transactions across both environments.
Regulatory and Liability Questions Must Be Addressed
Before AI agents can independently conduct large volumes of economic activity, regulators and financial institutions will need to establish clear rules.
Among the most important issues are:
- Whether an AI agent can be recognised as an authorised payment initiator.
- How delegated authority should be recorded and revoked.
- Who is liable for an agent’s mistake.
- How users can dispute agent-initiated transactions.
- How fraudulent or compromised agents should be identified.
- Whether transaction limits should vary by agent risk level.
- How consumer-protection requirements should apply to autonomous purchases.
- What evidence must be retained for regulatory and legal review.
Banks and payment service providers will also need mechanisms to distinguish between transactions initiated directly by humans and transactions initiated by AI agents.
Agent identity may eventually become as important as device identity is today.
Every transaction may need to carry information describing the responsible human, the executing agent, the applicable mandate and the authorisation system that approved the action.
India Has an Opportunity to Lead Agentic Payments
India already possesses several critical building blocks for an agentic payment economy:
- A mature real-time payment network.
- Large-scale digital identity infrastructure.
- Broad QR-code acceptance.
- Rapid fintech adoption.
- Regulated banks and payment providers.
- Emerging delegated-payment capabilities.
- A large digital consumer and merchant ecosystem.
The strategic opportunity is to combine these strengths into a trusted Agentic UPI framework. This would require cooperation between the Reserve Bank of India, the National Payments Corporation of India, banks, fintech companies, AI developers, cybersecurity platforms, merchants and consumer-protection authorities.
India could define global standards for how autonomous AI agents are authorised to spend money on behalf of humans and organisations. Such standards could include:
- Agent identity and registration.
- Machine-readable payment mandates.
- Delegated-authority credentials.
- Dynamic transaction limits.
- Human-in-the-loop controls.
- Tamper-evident payment evidence.
- Independent verification.
- Agent risk scoring.
- Emergency revocation and kill switches.
- Standardised dispute and refund mechanisms.
The Next Evolution of UPI
UPI transformed Indian payments by making bank-to-bank transfers simple, instant and interoperable. Its next evolution could be even more significant. UPI may move from enabling humans to make digital payments to enabling AI agents to safely participate in the economy under human-defined authority. The key transition will be from “a user authorises an individual payment” to “a user authorises a bounded economic objective.”
Once that transition occurs, AI agents could continuously purchase products and services while remaining constrained by spending limits, business policies, regulatory requirements and human judgment.
UPI as the settlement rail. Programmable mandates as the authority mechanism. AI agents as the execution interface. Cryptographic evidence as the trust layer.
UPI has the potential to become the most native payment interface for AI agents conducting domestic commerce in India.
Globally, the agentic economy will likely remain multi-rail. UPI may dominate regulated domestic fiat transactions, while stablecoins and other digital payment systems support cross-border and machine-to-machine commerce. But regardless of which payment rail is used, the decisive capability will not simply be the ability of an AI agent to pay. It will be the ability to prove that the agent was authorised to act, remained within its permitted boundaries and produced a verifiable record of every economic decision it made. That authorisation and accountability layer will determine whether billions of AI agents can participate safely in the global economy.
