For most of digital commerce, the customer has also been the operator.
People search, compare, visit websites, interpret offers, select payment methods and decide when to complete a transaction. Software supports each step, but the person remains present throughout the journey.
That boundary is beginning to move.
India is preparing a framework that would allow AI agents to make limited purchases through the Unified Payments Interface without requiring approval for every transaction. According to Reuters, the planned Unified Agent Protocol would initially support low-value, frequent purchases through spending limits, identity checks, audit trails and delegated authority.
Visa has introduced a protocol that helps merchants distinguish approved shopping agents from ordinary bots. Google has developed standards for agent-led discovery, checkout and payment. SAP is applying agents to procurement and spend management.
These initiatives are often described as improvements to checkout.
The larger change is that a new participant is entering the commercial relationship.
An agent may interpret what a person wants, search across suppliers, compare complete offers, evaluate restrictions and execute the purchase. The customer still determines the objective. The agent increasingly determines how the market is examined.
The customer owns the intent. The agent controls the path.
The next customer may never visit the website
The familiar digital funnel assumes that a business first earns attention.
A customer sees an advertisement, follows a recommendation, enters a marketplace or searches for a product. The merchant then controls the environment in which the choice is presented. Photography, page structure, copy, reviews, merchandising and checkout are designed to move the person toward a purchase.
An agent can bypass much of that sequence.
A person may ask for a carry-on suitcase that fits a specific airline, arrives before Friday, has a strong warranty and costs less than a fixed amount. The agent can translate that request into criteria, examine several sellers and return with one recommendation or complete the purchase within an approved limit.
The winning merchant may never receive a conventional visit.
Its product may be discovered through a platform, evaluated through structured data and purchased through an agent-compatible transaction layer. The customer may see only the final recommendation, confirmation or receipt.
This does not eliminate the customer relationship. It redistributes it.
The agent or platform may own the interface. The payment provider may verify authority. The merchant remains responsible for the product, fulfilment and post-purchase experience. Several companies can therefore participate in a sale without any one of them controlling the complete journey.
Businesses accustomed to optimizing pages for human attention will also need to optimize their commercial systems for machine interpretation.
Delegated purchasing requires a new contract
A person can reconsider a purchase at any moment. An autonomous agent needs a defined boundary before it acts.
That boundary may include a maximum price, permitted merchants, product requirements, delivery deadlines, geographic restrictions and a period during which the authority remains valid. It may also specify when the agent must return for confirmation.
Google's Agent Payments Protocol approaches this problem through verifiable mandates that create evidence of the user's intent and the authority delegated to an agent. India's proposed UPI framework is expected to draw on existing mechanisms that allow a primary account holder to delegate payment authority and reserve funds for defined uses.
The principle is more important than any individual protocol.
An agent cannot rely on the kind of ambiguous consent that digital commerce has often tolerated. If software is allowed to spend money, the merchant and payment network need to know who authorized it, what was authorized and whether the transaction remained inside those terms.
This turns permission into part of the product.
A useful transaction record may need to show the request the agent received, the options it considered, the rule under which it selected one, the price and terms presented at approval, and the credential used to pay.
Trust is no longer expressed only through the appearance of a familiar interface. It must also exist as evidence that can be inspected after the transaction.
Product data becomes part of the sales function
Human customers can compensate for imperfect commercial information.
They infer meaning from images, contact support when a policy is unclear and sometimes continue through a checkout even when the final cost appears late.
Agents are less forgiving.
To compare offers accurately, an agent needs current information about inventory, variants, delivery, taxes, fees, warranty, compatibility, subscriptions, cancellation and returns. It must know whether a product can actually reach the customer under the stated conditions.
This information often sits across several systems. Marketing owns the product page. Operations owns inventory. Finance determines price and tax. Logistics supplies delivery estimates. Customer support interprets exceptions. Legal writes the policy.
Agentic commerce forces those sources to agree.
Google's Universal Commerce Protocol is intended to provide a common language across discovery, buying and post-purchase support. The protocol reflects a broader commercial requirement: agents need more than a page they can scrape. They need data and actions that can be understood consistently across businesses.
For an agent, the catalogue is not collateral. It is the storefront.
A merchant with excellent products but stale availability, incomplete attributes or ambiguous return terms may become invisible at the point of comparison. A less distinctive competitor with reliable data may be easier for the agent to recommend and safer for it to transact with.
Product information therefore moves closer to revenue responsibility. Accuracy, structure and update speed become part of conversion rather than administrative work performed after merchandising.
Brand influence is compressed into evidence
Agents do not remove brand from commerce. They change where it exerts influence.
A customer may tell an agent to buy from a preferred company, preserve loyalty status or avoid unfamiliar sellers. Reputation can become a rule inside the request. In those cases, past brand investment shapes the agent's decision before comparison begins.
But when the request is open, the agent may evaluate the market through observable criteria.
Price, availability, verified reviews, warranty length, delivery reliability, product compatibility and return conditions can carry more weight than the environment in which they are presented. The agent may not experience the photography, tone or sequence that once differentiated the merchant at the point of selection.
Brand must therefore become more legible as evidence.
A claim of quality needs supporting attributes. A promise of convenience needs dependable fulfilment data. A reputation for service needs policies that are clear enough for an agent to evaluate before it buys.
The strongest brands will still matter. They may become even more valuable as customers use them to constrain what an agent is allowed to consider. Weaker forms of differentiation, however, may be compressed when several products satisfy the same machine-readable criteria.
The strategic task is to identify which parts of the brand can survive translation into the rules and evidence used by an intermediary.
Offers and negotiation become programmable
When the buyer can examine the market continuously, a posted price may become the beginning of the transaction rather than the complete offer.
An agent can compare bundles, delivery promises, financing, loyalty benefits and return conditions. It can wait for a threshold, request an eligible discount or decide whether a cheaper price compensates for slower fulfilment.
Google's early Universal Commerce Protocol work includes support for direct offers that can be presented to high-intent shoppers. The broader direction is toward commercial terms that software can evaluate and, within defined limits, negotiate.
This creates opportunity and risk.
Merchants can use inventory, margin and customer context to assemble more relevant offers. Customers can avoid the effort of monitoring prices or searching for codes. But poorly governed automation can produce inconsistent pricing, channel conflict or discounts that destroy the economics of the sale.
Offer logic therefore needs explicit boundaries.
Which terms may change? Which customers or agents are eligible? What minimum margin must be protected? Does an offer remain valid if inventory, delivery cost or risk changes? Can the business explain why two apparently similar transactions received different terms?
In agentic commerce, pricing rules may become as important to the customer experience as the price itself.
Merchants must recognize the buyer
For years, automated traffic has often been treated as a problem to block.
Bots scrape prices, consume inventory, overwhelm infrastructure and automate purchases of scarce products. A legitimate shopping agent may initially look very similar. It can visit quickly, request structured information and attempt to act without a person navigating each page.
Merchants need a way to distinguish useful automation from hostile automation.
Visa's Trusted Agent Protocol is designed for this boundary. It allows an approved agent to present signed credentials that identify the agent, link it to a consumer and support payment. A merchant or site-protection provider can verify those credentials before deciding what the agent may access or do.
Recognition does not require unrestricted access.
A merchant may allow a verified agent to retrieve current availability and complete a purchase while still applying rate limits, inventory controls and additional checks for unusual behaviour. High-risk categories may require confirmation even when routine purchases do not.
The result is a new access layer between public browsing and a private commercial relationship.
Agents will need identity. Merchants will need policy. Payment providers will need proof that the identity, customer authority and payment credential belong to the same transaction.
Bot management therefore begins to overlap with sales infrastructure.
Loyalty, attribution and customer ownership become contested
Digital businesses have learned to measure the path to purchase through impressions, clicks, visits and conversion.
An agent-led journey can remove several of those events.
The customer may express intent inside a general assistant. The assistant may compare products through a protocol, complete checkout without opening the merchant's interface and return only when a decision or confirmation is required.
The merchant still wins or loses the sale. Its conventional funnel may show little of what happened.
New measures become necessary:
- Whether the merchant's products were eligible for agent discovery.
- How often the merchant was included in a comparison.
- Which criteria caused an offer to be selected or rejected.
- How often authorization, payment or fulfilment failed.
- Whether agent-originated orders produce different margins, returns or repeat behaviour.
- Whether the customer can still be recognized across the agent, payment and merchant relationship.
Loyalty also needs to travel.
If a customer buys through an agent, the merchant should be able to recognize an existing membership, apply the correct benefits and preserve the purchase history where permission allows. If that identity remains trapped inside the agent platform, the merchant can become a fulfilment layer while the intermediary accumulates the customer relationship.
Attribution will therefore become partly contractual and partly technical. Platforms, payment providers and merchants will negotiate who receives customer data, who may communicate after purchase and which participant receives credit for creating the demand.
The businesses that ignore this layer may gain transactions while gradually losing the relationship those transactions once created.
Post-purchase service becomes part of the sale
An agent does not need to stop after payment.
It can track delivery, compare the item received with the promise made, request a refund, monitor a trial, cancel a subscription or act when a warranty condition is met. It can also remember policies that a person would normally forget.
This automates customer vigilance.
Commercial friction that once survived because the customer lacked time becomes easier to challenge. A short return window, an unexpected renewal or a price-protection promise can be monitored continuously.
Merchants will need machine-readable processes for cancellation, returns, exchanges, delivery exceptions and support escalation. The agent should be able to understand what happened and which action remains available without recreating the transaction through a support conversation.
This makes post-purchase design part of discoverability.
If two offers are otherwise similar, an agent may prefer the merchant whose fulfilment and service commitments are clearer and easier to enforce. A generous policy is valuable only if the system can recognize when it applies and execute it without unnecessary ambiguity.
The commercial journey becomes a loop. Product data supports discovery. Transaction data supports execution. Service data determines whether the promise was kept and influences the next selection.
Bounded transactions will arrive first
The first useful agentic purchases are unlikely to involve unlimited discretion.
India's proposed framework is expected to begin with low-value, frequent purchases. The logic is practical. Groceries and routine items can be governed by clear limits, familiar merchants and reversible decisions. The value comes from removing repetition rather than delegating a rare, consequential choice.
Business procurement may develop along a similar path.
An organization already has approved suppliers, budgets, product categories and authorization rules. An agent can search within those boundaries, prepare an order, compare terms and execute when the transaction meets policy. SAP describes this direction as autonomous spend management, applying agents across sourcing, supplier management, receiving and invoicing.
The operating environment is constrained, but that is what makes automation credible.
A useful progression is likely to be gradual:
- Observe transactions and identify recurring decisions.
- Recommend a purchase while the customer remains in control.
- Prepare the order and request confirmation.
- Execute routine transactions inside explicit limits.
- Expand authority only when the evidence supports it.
The important question is not whether an agent can buy something.
It is whether the complete system can prove that the purchase reflected the customer's intent, used accurate information and remained inside an agreed boundary.
Businesses need an agent-readable commercial layer
Agentic commerce will not require every company to rebuild its business around a new protocol immediately.
It does require a different readiness question.
Can software acting for a customer understand what the company sells, determine whether the offer is actually available, evaluate the complete terms, prove its authority and complete the transaction safely?
For many businesses, preparation will begin with fundamentals:
- Make product, inventory, delivery and policy data consistent across systems.
- Expose the full cost and conditions of an offer before authorization.
- Define which actions an agent may take and when confirmation is required.
- Authenticate legitimate agents without weakening protection against hostile automation.
- Preserve customer recognition, loyalty and post-purchase rights across channels.
- Measure agent discovery, selection, completion, exceptions and long-term customer value.
These are not separate technical tasks. Together, they form the commercial surface that an agent encounters.
At BraidNest, we view agentic commerce as a product and operating-model shift rather than a checkout feature. The work spans product data, permissions, payments, fulfilment, customer ownership and the systems through which those decisions are executed.
Companies should not begin by asking how to add an agent to the storefront.
They should ask whether the business is prepared for software to arrive on behalf of the customer.
The next competitive advantage may not be having an agent of your own. It may be becoming the business other agents can understand, trust and transact with.


