On 1 September 2026, Inc42 reported - citing Reuters - that NPCI plans a Unified Agent Protocol letting AI agents make UPI payments inside pre-set limits, without a human approving every transaction. NPCI declined to comment. Eight months earlier Google and Shopify published the Universal Commerce Protocol, and by 24 March 2026 Shopify had switched on Agentic Storefronts, surfacing merchant catalogues inside ChatGPT, Microsoft Copilot, AI Mode in Google Search and the Gemini app. If an agent cannot read your catalogue, it cannot recommend you - and your competitor's structured feed wins by default.
What agentic commerce actually means for an Indian store
Agentic commerce is shopping where an AI assistant does the discovery, comparison and sometimes the checkout on the buyer's behalf. Shopify's own description is blunt: agents help shoppers discover, compare and purchase inside a conversation instead of browsing a storefront. That changes who your customer is. Your catalogue is no longer read by a human scanning product cards - it is parsed by software that wants titles, prices, stock status, shipping rules and return policies in machine-readable form, and silently skips anything it cannot resolve. For a D2C brand in Hyderabad or a retailer in Bengaluru, catalogue hygiene has become a distribution problem rather than a housekeeping chore.
- •Discovery moves from your product page to a conversation you do not control, so your data has to travel to the agent.
- •Agents resolve facts, not vibes - an unpriced variant or a stale in-stock flag is simply dropped from consideration.
- •Shopify says its agentic storefronts package products, checkout and brand information so AI platforms can present them natively.
- •The same structured data that feeds agents also feeds Google's merchant listings, so the work is not wasted if agentic buying stalls.
- •None of this requires Shopify - it requires a clean feed, valid markup and an honest availability signal.
What NPCI has reportedly proposed - and what is actually confirmed
Separate the two carefully, because the difference matters when you are budgeting. Reported: Inc42, relaying Reuters, said on 1 September 2026 that NPCI would unveil a Unified Agent Protocol at the 2026 Global Fintech Fest in Mumbai, letting AI agents transact on UPI within limits the user sets in advance, starting with low-value, high-frequency purchases such as grocery orders. The mechanics lean on two things UPI already has - UPI Circle, which lets a primary user delegate spending authority, and Reserve Pay, which blocks funds for future payments. September 2026 press reports also describe a registry of verified agents inside that protocol. None of it is an NPCI announcement: NPCI declined to comment to Inc42, and no framework has been published, including who carries the loss when an agent misfires.
- •Confirmed and already public: on 10 October 2025 Razorpay, NPCI and OpenAI announced agentic payments on ChatGPT as a pilot.
- •That pilot named Axis Bank and Airtel Payments Bank as banking partners and BigBasket as the first merchant to accept agentic payments.
- •It runs on UPI Circle and UPI Reserve Pay, with a one-time authorisation, an upfront spending cap, real-time tracking and instant revocation.
- •Reported but unconfirmed: the Unified Agent Protocol itself, the agent registry, the launch timing and the liability rules.
- •Treat the payment layer as a dependency you do not control, and build the parts you do - catalogue, feed, policies, checkout - now.
How big is the AI shopping channel already?
Big enough that waiting is a decision. Shopify published Q1 2026 figures showing AI-driven traffic to its stores grew eightfold year over year and orders from AI-powered searches rose nearly thirteenfold, with new buyers ordering through AI channels at nearly twice the rate of other channels. Anthropic, announcing its commerce-agent blueprint on 2 September 2026, said retailers running shopping agents on Claude report carts up to 35 percent larger and shoppers 60 percent more likely to complete a purchase. Read those as directional vendor numbers, not audited market data. On the India side, the rail these agents would eventually touch is not small: UPI processed 24.51 billion transactions worth Rs 29.82 lakh crore in August 2026, averaging 791 million a day across 752 live member banks.
UCP vs ACP vs Agentic Storefronts vs NPCI's UAP: which protocol matters to you?
Several things get called agentic commerce and they do different jobs. UCP and ACP are open protocols for how an agent talks to a shop. Shopify's Agentic Storefronts is a product that implements the plumbing for you. NPCI's reported Unified Agent Protocol is a payment-authority layer specific to India. You do not have to pick a winner - the underlying requirement, a clean machine-readable catalogue with trustworthy prices and stock, is identical across all of them.
| Protocol or product | Status in September 2026 | Backed by | What it covers | India relevance |
|---|---|---|---|---|
| Universal Commerce Protocol (UCP) | Published spec at ucp.dev, Apache 2.0 | Google and Shopify, with Amazon, Microsoft, Meta, Stripe, Walmart, Target, Etsy | Manifest at /.well-known/ucp, capability negotiation, catalogue search, cart, OAuth 2.0 identity linking, checkout, order management | Flipkart endorses it; no India-specific payment rail in the spec |
| Agentic Commerce Protocol (ACP) | Open source, Apache 2.0 | OpenAI and Stripe | Product feeds by API or SFTP, an Agentic Checkout Spec for merchants, a Delegated Payment Spec for payment providers; REST and MCP | The route into ChatGPT shopping, where the October 2025 UPI pilot ran |
| Shopify Agentic Storefronts | Live to eligible merchants from the week of 24 March 2026 | Shopify | Syndicates catalogue, checkout and brand data to ChatGPT, Microsoft Copilot, AI Mode in Google Search and the Gemini app | For Shopify merchants; the Agentic plan extends Shopify Catalog to others |
| NPCI Unified Agent Protocol (UAP) | Reported, not confirmed - NPCI declined to comment to Inc42 | NPCI (reported); pilot involves Razorpay, OpenAI, Axis Bank, Airtel Payments Bank, BigBasket | Reported to let agents pay on UPI within pre-set limits via UPI Circle and Reserve Pay, starting with low-value purchases | India-only, and decides whether agentic checkout works natively on UPI |
| Claude commerce agents blueprint | Published 2 September 2026 on GitHub | Anthropic, with Shopify, Visa, Mastercard, Zomato and Square | Reference code for shopping and merchant agents on the Claude API, Amazon Bedrock, Microsoft Foundry and Vertex AI | Zomato is a named partner, so Indian food ordering is already in scope |
Your 90-day agent-readiness checklist
No replatform needed - just a product-data owner, a fortnight of cleanup and discipline about freshness. Work in order; the feed unlocks everything else.
Make your store agent-ready in 90 days
Product data: the schema and feed fields agents actually read
Google's guidance is that supplying both structured data on the page and a Merchant Center feed maximises your eligibility and helps Google verify what you sell. Treat the two as one dataset with two publication channels. The required Merchant Center attributes are id, title, description, link, image_link, availability and price - id capped at 50 characters, title at 150, description at 5,000, price in ISO 4217 form. Availability must be in_stock, out_of_stock, preorder or backorder, and must agree with your landing page and checkout. On the page, merchant listing markup wants price, priceCurrency, availability and priceValidUntil, plus shippingDetails and MerchantReturnPolicy. Mismatch between feed, markup and checkout total is the commonest reason a product gets filtered out.
- •GTIN is strongly recommended - Google states the correct GTIN gives the best user experience and the best performance.
- •brand is required for new products except movies, books and recordings; mpn only if the product has no manufacturer-assigned GTIN.
- •item_group_id is required for variants in several markets and for all free-listing variants - use it even where it is optional.
- •maximum_retail_price is available only in India: submit MRP in INR and keep it consistent with the landing page and checkout.
- •priceValidUntil stops an agent quoting a festive-sale price in November that you retired in October.
Checkout, UPI and payments when the buyer is a bot
Agentic checkout splits into two questions: can the agent complete a purchase, and who authorised the money. ACP answers the first with an Agentic Checkout Spec for merchants and a Delegated Payment Spec for payment providers. UCP answers it with OAuth 2.0 identity linking and a checkout capability, while the merchant stays merchant of record. India's answer to the second question is still in flight, and the October 2025 pilot shows the shape to build around: a one-time authorisation, a spending cap set upfront, live tracking and instant revocation.
Expose a reliable price-and-stock endpoint
Agents re-check before they commit. A slow or occasionally wrong stock call means a cart abandoned by software that never complains.
Support a capped, revocable mandate
Ask your payment provider how it handles UPI Circle delegation and Reserve Pay fund blocking today, and what happens when a user revokes mid-order.
Decide your refund posture in writing
Agree what happens when an agent orders the wrong variant - who refunds, how fast, and whether the order is reversed before dispatch.
Separate fraud rules for agent traffic
Velocity checks tuned for humans will flag a legitimate agent placing five small orders in a minute. Tune thresholds instead of blanket-blocking.
Keep the GST invoice path intact
An agent-placed order still needs a compliant GST invoice with the right place of supply. No new channel should bypass your invoicing rules.
Test the unhappy path
Out of stock after authorisation, partial fulfilment, address failure - script these before an agent finds them in production.
What NOT to do yet
There is noise in this category and some of it will cost you money. These are the errors worth naming.
⚠️Replatforming your store to chase agentic commerce
Consequence: Six months and a large budget on migration while the real blocker - dirty product data - travels to the new platform unchanged.
Solution: Fix the catalogue, feed and markup on what you already run. Every protocol in play reads the same fields.
⚠️Assuming NPCI's Unified Agent Protocol is live and dated
Consequence: Roadmaps built on a press report NPCI itself declined to confirm, with no published liability framework.
Solution: Plan for the confirmed pilot shape - capped, revocable UPI mandates - and revisit when NPCI publishes the framework.
⚠️Blocking all bot user agents at the CDN
Consequence: The shopping agents you want are refused alongside the scrapers you do not, and your products quietly vanish from AI surfaces.
Solution: Allow-list the commerce agents you want, rate-limit rather than ban, and monitor what each crawler requests.
⚠️Publishing a price in markup that differs from checkout
Consequence: Products get filtered from merchant listings and agents lose trust in your feed - the hardest thing to win back.
Solution: Make one system the source of truth and generate the feed, the page markup and the checkout total from it.
⚠️Buying an agentic commerce audit before doing the basics
Consequence: A report telling you to add GTINs and fix availability, for the price of a quarter of engineering time.
Solution: Run the 90-day checklist first, then buy advice only for whatever remains genuinely unclear.
Measuring agent traffic before your competitors do
You cannot argue for budget on a channel you are not counting. Most Indian storefronts bucket AI-referred sessions into direct or other, so agent-assisted revenue is invisible just as it starts mattering. Shopify reports order attribution back into merchant admin for agentic orders; the same principle applies wherever you sell. Capture the referrer, tag it, and report it separately from week one.
- •Build a named AI-channel segment in analytics, covering assistant referrers and any agent-specific parameters your platform passes through.
- •Track share of catalogue visible to agents - how many SKUs pass feed validation - as a real operational metric.
- •Watch feed disapproval reasons weekly; they are the cheapest early-warning system you have.
- •Measure price and stock freshness in minutes, and treat a stale field as an incident rather than a backlog item.
How Tech Arion helps with agentic commerce readiness
We build and fix the layer agents actually read. Our e-commerce team at techarion.com/services/ecommerce-development builds stores with clean variant structures, Razorpay and UPI checkout, WhatsApp ordering and a single catalogue source feeding the website, the product feed and the chat channel together - so one price change lands everywhere at once. For restaurants and retail counters, techarion.com/services/qr-solutions keeps QR menus reading from that same catalogue rather than a PDF nobody updates. And techarion.com/services/analytics-reporting sets up the AI-channel segmentation, feed-health monitoring and attribution reporting that turn agent traffic into a line in your monthly numbers. Most engagements start small: a catalogue and feed audit against Merchant Center requirements and schema.org Product and Offer markup, a fix list ranked by revenue impact, then the build. We are a Hyderabad company working across India, so the scope always includes the India-specific details - MRP, GST invoicing and UPI - that global checklists leave out.
Frequently asked questions about agentic commerce in India
The questions Indian founders and retail CTOs are asking us this month.
Frequently Asked Questions
Make your store readable by AI shopping agents
Agentic commerce rewards the boring work: clean variants, honest stock, matching prices and machine-readable policies. Tech Arion audits your catalogue against Google Merchant Center requirements and schema.org Product and Offer markup, fixes what blocks you, and builds the checkout, WhatsApp ordering and analytics around it. We are based in Hyderabad and we cover the India-specific details - MRP, GST invoicing and UPI - that global agentic-commerce checklists skip. Send us your catalogue size and current platform and we will tell you what a 90-day readiness plan looks like for you.
Sources & References
Primary sources fetched on 24 September 2026:
- 1.
Google Developers Blog. (11 Jan 2026). Under the Hood: Universal Commerce Protocol - discovery via a JSON manifest at /.well-known/ucp, capability negotiation, checkout and post-purchase; built on A2A, AP2 and MCP; spec on GitHub under Apache 2.0.
View Source - 2.
Universal Commerce Protocol. ucp.dev - co-developed by Google, Shopify, Amazon, Microsoft, Meta, Stripe and others; endorsing partners include Flipkart; capabilities cover catalogue search, cart building, OAuth 2.0 identity linking, checkout and order management; Apache 2.0.
View Source - 3.
Shopify. (24 Mar 2026). Agentic commerce momentum - Agentic Storefronts live for merchants across ChatGPT, Microsoft Copilot, AI Mode in Google Search and the Gemini app; Agentic plan opened to non-Shopify merchants via Shopify Catalog.
View Source - 4.
Shopify. (18 Jun 2026). How agentic commerce works - Q1 2026: AI-driven traffic to Shopify stores grew 8x year over year, orders from AI-powered searches rose nearly 13x, and new buyers ordered through AI channels at nearly twice the rate of other channels.
View Source - 5.
Agentic Commerce Protocol. agenticcommerce.dev - an open standard for commerce flows between buyers, AI agents and businesses, built by Stripe and OpenAI, open source under Apache 2.0, REST and MCP compatible.
View Source - 6.
OpenAI. developers.openai.com/commerce - ACP integration via product feeds submitted by API or SFTP, the Agentic Checkout Spec for merchants and the Delegated Payment Spec for payment service providers.
View Source - 7.
Anthropic. (2 Sep 2026). Building commerce agents with Claude - blueprint published at anthropics/commerce-agents; partners include Shopify, Visa, Mastercard, Zomato and Square; retailers report carts up to 35% larger and shoppers 60% more likely to complete purchases.
View Source - 8.
Google. Merchant Center product data specification - required attributes id, title, description, link, image_link, availability and price; GTIN, brand, MPN and item_group_id guidance; maximum_retail_price available only in India; 500 x 500 px image minimum enforced from 31 January 2027.
View Source - 9.
Google Search Central. Product structured data - merchant listing and product snippet markup, price, priceCurrency, availability, priceValidUntil, shippingDetails and MerchantReturnPolicy; providing both structured data and a Merchant Center feed maximises eligibility.
View Source - 10.
Inc42. (1 Sep 2026). NPCI to launch agentic payments on UPI: report - citing Reuters, NPCI plans a Unified Agent Protocol using UPI Circle and Reserve Pay for low-value, high-frequency payments; NPCI declined to comment.
View Source - 11.
MediaNama. (10 Oct 2025). Razorpay, NPCI and OpenAI launch agentic payments on ChatGPT - pilot with Axis Bank and Airtel Payments Bank as banking partners and BigBasket as the first merchant, via UPI Circle and UPI Reserve Pay with a one-time authorisation and spending cap.
View Source - 12.
MediaNama. (17 Sep 2026). UPI August 2026 data - 24.51 billion transactions worth Rs 29.82 lakh crore, 791 million transactions a day, average ticket about Rs 1,217, across 752 live member banks.
View Source