What Walmart's Agent Gateway signals for agentic commerce
Agentic commerce is shopping in which an AI agent searches, compares and buys on a customer’s behalf, within limits the customer sets. On 21 October 2026, Walmart announced Agent Gateway, an MCP server through which approved AI agents can search Walmart and Sam’s Club inventory and buy for customers. Walmart also announced Agent Profiles, which will let shoppers set spending limits and approval rules for each agent. It also published terms that agents must follow or be blocked (TechCrunch). The gateway starts as a beta for companies and developers, with Agent Profiles due to follow.
The timing is telling. Amazon has been blocking third-party shopping agents, including Meta’s Muse and Perplexity’s agent, from buying on Amazon.com. Two of the world’s largest retailers have taken opposite positions on AI shopping agents in the same month.
This blog explains what that split means for Indian e-commerce, what makes a catalogue agent-ready, and how to build a governed agentic commerce channel on AWS.
Why agentic commerce matters for Indian e-commerce now
India is moving quickly. Google began testing purchases from Flipkart inside the Gemini app and AI Mode in Search in India (TechCrunch). On payments, Reuters reported in September 2026 that India is preparing a framework that would let AI agents make small UPI payments without approval for every transaction (Reuters via The Star).
For marketplaces such as Flipkart and Amazon India, fashion platforms such as Myntra and Ajio, quick-commerce apps and D2C brands, this raises a practical question. When a customer asks an AI assistant to find a cotton kurta under ₹1,500 that arrives by Friday, will the agent find your product, trust your price and stock data, and complete the order?
Agentic commerce does not replace your app or website. It adds a channel where the visitor is software, and software relies on structured data more than banners.
Open, block or partner: how retailers are responding
| Approach | Example | Upside | Risk |
|---|---|---|---|
| Block outside agents | Amazon.com | Keeps control of the customer relationship and on-site ads | Misses agent-driven demand; unidentified bots may still scrape |
| Open with rules | Walmart Agent Gateway | Reaches agent shoppers through controlled, identified access | Needs clean product data and governance |
| Partner with an AI platform | Flipkart test with Google Gemini | Fast reach through a major assistant | Depends on the platform's terms and data sharing |
| Wait and watch | Not applicable | No spend today | Competitors and platforms set the standards first |
Most Indian retailers exploring agentic commerce will end up with a mix: a governed channel for the agents and platforms they trust, and firm bot controls for everything unidentified.
What makes a catalogue ready for agentic commerce
Multimodal AI agents can interpret images, but dependable product comparison and transactions still require structured, machine-readable data. Five foundations decide whether your products show up and convert when an AI agent does the shopping:
- Structured product data. Attributes such as size, fabric, fit, compatibility and care are complete and consistent, not buried in images or marketing copy. For fashion, size charts need to exist as data, not as a picture.
- Real-time price and inventory. An agent that buys a sold-out item or uses a stale price creates cancellations and returns. Delivery promises should be accurate at the pincode level.
- A governed access layer. An MCP server or API exposes search, product detail, cart and order tools to approved agents, with agent identity, scopes and rate limits.
- Customer consent and limits. Shoppers control which agents can act for them, with spending caps and approval rules, similar to Walmart’s Agent Profiles.
- Channel measurement. Agent traffic is tagged as its own channel, so you can see its conversion, order value, returns and cancellations.
Most retailers already hold this data. The work is making it complete, current and safely accessible.
Reference architecture for agentic commerce on AWS
A governed agentic commerce channel can be built from managed AWS services around your existing commerce platform:
| Layer | AWS services | Role |
|---|---|---|
| Catalogue data | Amazon S3, AWS Glue | Collect, clean and join product, price and inventory data |
| Attribute enrichment | Amazon Bedrock | Draft missing attributes and agent-friendly descriptions from existing content, with human review before publishing |
| Product search | Amazon OpenSearch Service | Keyword and semantic (vector) search that agents can query |
| Live price and stock | Amazon DynamoDB, Amazon EventBridge | Keep price, stock and delivery data current as it changes |
| Agent access | Amazon Bedrock AgentCore Gateway, Amazon API Gateway | Expose existing APIs as MCP-compatible tools with identity and access controls |
| Bot control | AWS WAF Bot Control with Web Bot Authentication | Verify signed AI agents, classify automated traffic, and block or rate-limit unidentified activity |
| Measurement | Amazon CloudWatch, Amazon Athena, Amazon QuickSight | Monitor the agent channel and report its performance |
Payments, order management and returns stay in your existing systems and payment gateway. The agent access layer sits in front of them rather than replacing them.
For implementation details, see the AWS documentation for Amazon Bedrock AgentCore Gateway and AWS WAF Bot Control.
How to measure the agentic commerce channel
| Metric | What it tells you |
|---|---|
| Catalogue coverage (%) | Share of SKUs with every required attribute filled |
| Agent-initiated sessions | How much demand arrives through AI agents |
| Agent conversion rate (%) | Whether agents trust and buy your products |
| Cancellations from stale data (%) | Accuracy of price, stock and delivery data |
| Agent-channel return rate (%) | Whether agent purchases match what customers wanted |
| Unidentified bot traffic | How much automated traffic sits outside your governed channel |
Start with catalogue coverage. It is the metric you can improve before any agent traffic arrives.
A phased roadmap for agent-enabled retail
- Audit. Score attribute completeness and data freshness for your top categories.
- Fix the data. Enrich missing attributes with reviewed AI help, and move price and stock to real-time updates.
- Pilot read-only access. Expose search and product-detail tools to one or two trusted agents or platforms.
- Add cart and checkout. Introduce ordering with consent controls, spending limits and your payment partner, in line with UPI’s agentic payment rules as they are finalised.
- Scale and measure. Track the agent channel against your other channels and widen access as results hold.
How Rapyder helps
Rapyder is an AWS Premier Tier Services Partner with the AWS Generative AI Competency. Our agentic AI services, data analytics and retail and e-commerce teams help retailers prepare product data, build governed MCP and API access layers, and measure AI-led shopping as a channel on AWS.
If you want to know where your catalogue stands, talk to Rapyder about an agentic commerce readiness assessment.
Can AI shopping agents find and trust your products?
Get an agentic commerce readiness assessment for your catalogue on AWS.
Frequently Asked Questions
Agentic commerce is a shopping model in which AI agents search, compare and purchase products for customers within defined permissions. The agent uses retailer data and tools to check products, prices, stock and delivery. Customers remain in control through approval rules and spending limits.
Walmart announced Agent Gateway on 21 October 2026. The MCP server lets approved AI agents search Walmart and Sam’s Club inventory and buy for customers. Walmart also announced Agent Profiles for agent-specific spending limits, approvals and operating rules. The announcement shows how agentic commerce is moving from experimental shopping assistants to governed retail infrastructure.
Yes, but agentic commerce in India is still at an early stage. Google has tested purchases from Flipkart inside Gemini and AI Mode, while India is reportedly preparing a framework for small UPI payments by AI agents. Retailers can prepare now by improving product data, API access and transaction controls.
MCP, or Model Context Protocol, is an open standard that lets AI agents connect to external tools and data through a consistent interface. For retailers, an MCP server can expose approved catalogue, search, cart and ordering functions. This provides a governed alternative to agents scraping public web pages.
No. Retailers should not block every AI shopping agent by default. A better approach is to allow verified agents through governed APIs or MCP tools while restricting unidentified automated traffic. This protects retailer systems without cutting off a potential source of demand.
Your catalogue is ready when key products have complete structured attributes, current prices and inventory, and reliable delivery information. Test how consistently agents can find, compare and select products using that data. Catalogue coverage is the best first readiness metric.
Agentic commerce on AWS can use Amazon S3 and AWS Glue to organize catalogue data, Amazon Bedrock for attribute enrichment, and Amazon OpenSearch Service for product search. Amazon Bedrock AgentCore Gateway can expose MCP-compatible tools, while AWS WAF helps verify and control automated traffic.
Retailers can start with one controlled, read-only agentic commerce use case, such as catalogue search or product comparison. First improve data completeness, inventory accuracy and API access. Add cart and checkout only after identity, consent, spending limits and monitoring are in place.
