Who Gave the Bot Permission? 

Pine Labs, UPI, and the rules that could decide how AI spends money 

The first time Aditi’s AI assistant tried to buy something on its own, she did not think - That is impressive. She thought - Who gave it permission? 

That is the real question behind the new wave of agentic payments. Pine Labs has recently introduced India’s first agent-to-agent payment capability by extending UPI’s standing mandate framework so AI systems can initiate payments autonomously. The payment still runs on existing UPI rails. What is new is the layer above it: the rules, permissions, and trust controls that decide when a machine is allowed to act.  

The rail is not the revolution. The rules are. 

Aditi runs procurement for a manufacturing company in Bengaluru. Her AI assistant watches stock levels, compares suppliers, checks delivery windows, and suggests when to reorder a part. That part is easy to understand. The hard part is not the shopping. It is the spending. How much can the agent use? Which suppliers are allowed? Does it need approval every time, or only for large orders? Can the permission be revoked? Can every action be traced later? 

That is where Pine Labs’ move matters. It suggests that the next frontier in UPI may not be just moving money faster. It may be teaching software how to use UPI safely on someone’s behalf. If a framework like this is eventually standardized by NPCI—potentially through initiatives such as its Unified Agent Protocol (UAP)—autonomous payments could stop looking like a proprietary add-on and start looking like a native part of UPI itself. 

This Is Not Just an Indian Story 

The idea of AI agents making purchases on behalf of people is emerging around the world, although each market is approaching it differently. 

In the United States and Europe, consumers are already beginning to experiment with AI assistants that can handle shopping tasks from start to finish. Instead of searching dozens of websites themselves, they might simply ask: 

"Find and book the cheapest nonstop flight to London for under $600 with free cancellation." 

Or: 

"Find me the best price on a premium coffee machine, compare reviews, choose the most reliable seller, and place the order." 

The AI agent searches multiple retailers, compares prices, delivery times, cancellation policies, customer reviews, and payment options before completing the purchase on the customer's behalf. 

Businesses are moving even faster. 

Large enterprises are increasingly using autonomous procurement agents to negotiate with suppliers, compare quotations, recommend vendors, and even conclude routine purchasing agreements within pre-approved limits. Companies such as Walmart have experimented with AI-powered negotiation platforms like U.S. based Pactum to automate parts of supplier negotiations, reducing weeks of procurement discussions to minutes. 

Retailers are also introducing their own shopping assistants. Amazon's Alexa, for example, can already reorder frequently purchased household items with minimal customer involvement. Future generations of these assistants are expected to make increasingly sophisticated purchasing decisions while staying within customer-defined preferences and spending limits. 

Different technologies. Different markets. The same destination. 

Software is slowly evolving from helping people shop to helping people buy. 

Sameer, who works on payment risk and controls, sees the story very differently. He does not think first about what the bot can buy. He thinks about what the bot must never do. He thinks limits, approvals, audit trails, merchant restrictions, and emergency stops. In his world, an AI agent is only useful if it can be trusted to stay inside the lines. 

The future of agentic payments may depend less on intelligence and more on permissioning. 

That is why the broader industry conversation is suddenly getting louder. If AI agents are going to transact across the internet, the system must be able to tell a real, authorized agent from a random bot. That is the direction Akamai’s recently described agentic security framework points toward: identity verification, observability, trust signals, and edge enforcement brought together so companies can distinguish human users, legitimate agents, and malicious traffic. The idea is simple to say and hard to do - know which bot is allowed to act, and which one is not. 

This is no longer a side issue. Across the industry, the trust layer is becoming the main event. Visa has integrated its payment network into ChatGPT so AI agents can shop and pay for users, while still using safeguards such as spending limits, approvals, and merchant restrictions. Mastercard has also expanded its Agent Pay work into the UAE. These moves all point in the same direction: commerce is becoming agentic, but it still needs rules. 

Cinthia, a security architect in the United States, thinks about that problem from the opposite end of the pipeline. She is not wondering whether the payment can be made. She is wondering whether the agent should be trusted at all. Who created it? Who owns it? Which human did it represent? Is it speaking for a legitimate customer, or pretending to? That is the heart of the “Know Your Agent” idea - a future where an AI agent is not just a line of code, but an identity that can be verified and governed. 

Before a bot can spend, the ecosystem must know who the bot is. 

That same question is now appearing in the global internet payments conversation. x402 revives the old HTTP 402 “Payment Required” idea for web-native transactions, aiming to let software, APIs, and agents pay directly inside normal internet requests. It is an elegant idea, but researchers have found that important questions still need answers: How do you know an agent is genuinely authorized to make a payment? How do you prevent someone from reusing an old payment request? And how do you ensure that transactions happen exactly as intended? In plain English: the idea is powerful, but the guardrails are still being built. 

Leena, who runs a digital platform business in London, sees why x402 matters. She sells access to APIs and services. Today, humans subscribe, log in, and pay. Tomorrow, software may do that itself. A machine could request a service, get a price, authorize payment, and continue working without waiting for a person to click through every step. That is exactly why protocols like Google’s Universal Commerce Protocol and other agentic-commerce standards are gaining attention: they are trying to make shopping, checkout, and post-purchase support work smoothly for AI-driven systems. 

Agentic commerce is no longer asking whether machines can buy. It is asking who gets to let them buy. 

Businesses can already see the attraction. Many purchases are repetitive, predictable, and governed by clear rules. Inventory replenishment, subscription renewals, supplier payments, and routine procurement decisions consume enormous amounts of time despite requiring little human judgment. If AI agents can handle these transactions within approved limits, businesses could reduce manual work, accelerate decision-making, and focus people on exceptions rather than routine tasks. 

The promise of agentic commerce is not novelty. It is less friction. 

But the risk is just as real. A recent security survey of autonomous commerce warns that the attack surface spans agent integrity, transaction authorization, inter-agent trust, market manipulation, and regulatory compliance. In other words, this is not only a payments problem. It is a systems problem. It touches identity, fraud, settlement, merchant trust, and governance all at once. 

That is why the Pine Labs story matters so much. It is not just a clever UPI feature. It is a practical step toward a future where autonomous software can spend within clearly defined human rules. The payment itself is still familiar. UPI still does the moving. The change is in what sits above it: the authorization and trust layer that decides who may act, how much they may spend, and when the permission ends. 

UPI still moves the money. The new question is who gets to ask it to move. 

That is also where lies an opportunity. The opportunity is not to change UPI, it is to build the layers around it: a platform that lets customers safely delegate payment authority to AI agents, an identity and trust framework that verifies agents and merchants, APIs that let agents and merchants discover and execute payments, and merchant-facing tools that accept AI-initiated transactions without creating chaos. The real value sits in the orchestration, control, and governance around the payment, not in the payment rail itself. 

And that is why this story feels bigger than one announcement. 

Pine Labs is showing what agentic payments could look like in India. 

Akamai and others are showing what trust infrastructure will need to look like on the open internet. 

Visa, Mastercard, x402, and UCP are showing that the global ecosystem is converging on the same unanswered question. 

How do we build a world where AI can spend, but only inside the boundaries humans set? 

The answer may define the next decade of payments. 

Not a wallet for the bot.  

A rulebook for the bot.