Does ChatGPT Need Amazon or Flipkart? The Inventory Infrastructure Question

agentic commerce India · AI agent buyer · autonomous AI commerce · ChatGPT buy products · AI inventory access

The Question Nobody Is Asking Clearly Enough

When an AI agent receives the instruction “buy me the best moisturiser under ₹800 that ships in two days” — what happens next?

The agent does not open a browser. It does not visit websites. It does not scroll through search results. It accesses product data from sources it is connected to, evaluates options against the user’s criteria, and completes a transaction through a pre-authorised payment mechanism.

The question is: which sources?

This is the inventory infrastructure question. And it has a direct answer — one that most discussions about Agentic Commerce quietly avoid because the answer is uncomfortable for anyone who has spent years building brand independence from aggregator platforms.

“The agent does not prefer aggregators. It goes where inventory is legible, the transaction is possible, and the trust is established. Right now, that is most reliably an aggregator. It does not have to stay that way.” — Anurag Gupta, Founder, ShodhDynamics.com

What Inventory Legibility Actually Means

An AI agent can only evaluate what it can read.

Not read in the human sense — not browse a website and interpret marketing copy. Read in the machine sense — access structured product data with precise attributes, real-time availability, accurate pricing, and unambiguous identity.

This is what inventory legibility means: the condition of a product catalogue being structured, current, and accessible enough for an AI agent to evaluate it accurately.

Inventory legibility has three dimensions:

Structure — product data expressed in machine-readable form. Not “perfect for gifting” — but category, material, dimensions, weight, price, variant options, and availability declared as explicit attributes. The difference between a product description written for a human and one structured for an agent is not style — it is the difference between data an agent can evaluate and noise it cannot process.

Currency — stock status, pricing, and availability accurate in real time. An agent completing a transaction on a user’s behalf cannot afford to select a product that is out of stock or priced differently than declared. OpenAI’s product feed specification for ChatGPT Shopping allows merchants to refresh catalogue data every 15 minutes. A brand updating its catalogue daily is operating on stale data in a system built for 15-minute refresh cycles. That gap is the difference between a completed transaction and a failed one.

Accessibility — product data reachable by the agent through a direct connection. Not via a webpage a human would browse. Via an API endpoint, a structured feed, or a direct catalogue integration that the agent’s platform can query at the moment of the transaction.

When all three dimensions are present — the product is inside the agent’s agentic consideration set: the subset of products an AI agent will actively evaluate for a given query.

When any one is missing — the product is invisible to the agent regardless of how good it is, how well-known the brand is, or how much the user might prefer it if they were making the decision themselves.

One further constraint shapes the agentic consideration set in practice: ChatGPT surfaces up to 36 curated product recommendations per shopping query. The brands that appear in those 36 are the ones with structured, accessible, current product data. The brands outside that 36 do not exist in that transaction — regardless of how well-known they are to human buyers.

Why Aggregators Currently Win — And Why It Is Not Permanent

Amazon and Flipkart did not set out to be Agentic Commerce infrastructure. But they built it by accident — as a consequence of solving a different problem.

When Amazon built its product catalogue for human search and browse, it created something else simultaneously: a machine-readable product database with consistent attribute structure, real-time inventory sync, and API accessibility. Every product on Amazon has a category, a title following consistent conventions, structured attributes (colour, size, material, weight), live stock status, and a price that updates dynamically.

This is exactly what inventory legibility requires. Amazon’s catalogue is not more AI-readable than a well-built independent brand catalogue because Amazon is smarter. It is more AI-readable because Amazon enforced attribute structure on sellers as a condition of listing. Sellers had no choice — fill in the required fields or the product cannot be listed.

The result: millions of products that are immediately accessible to AI agents, across a catalogue that an agent can navigate, filter, and evaluate without any additional integration work.

Flipkart, Meesho, and other Indian aggregators have the same structural advantage for the same structural reason.

The uncomfortable implication:

A D2C brand that spent years building direct customer relationships to escape aggregator dependence may find itself invisible to AI agents — while its competitor’s product, listed on Amazon with reluctance and thin margins, is the one getting recommended and purchased.

This is the aggregator advantage compounding into the agentic layer. Not because aggregators are better at AI. Because they accidentally built the infrastructure layer that AI agents need.

Why it is not permanent:

The aggregator advantage is a first-mover advantage in inventory legibility — not a permanent structural monopoly. A brand that builds machine-readable product data, real-time inventory access, and payment integration can participate in Agentic Commerce directly. The infrastructure required is buildable. The window to build it is open. What the aggregator has by default, an independent brand can build by intention.

The question is whether Indian D2C brands build that infrastructure before the agentic layer scales — or discover its absence when it is already too late to establish category positions.

“What the aggregator has by default, an independent brand can build by intention. The window to do so is open. The question is whether Indian brands build it before the agentic layer scales — or discover its absence when category positions are already occupied.” — Anurag Gupta, Founder, ShodhDynamics.com

The Three Sources AI Agents Actually Use

To understand the inventory infrastructure question precisely, it helps to map the sources an AI agent actually draws from when evaluating a purchase.

Source 1 — Connected aggregator catalogues The path of least resistance. Amazon, Flipkart, and marketplace catalogues are connected by default to many agentic platforms because they represent the highest transaction volume and the most consistently structured product data. An agent that can access an aggregator catalogue can immediately evaluate millions of products without any additional integration.

Source 2 — Direct brand integrations Brands that have built direct API connections with agentic platforms — or whose e-commerce platforms (Shopify, WooCommerce with proper configuration) expose structured product feeds — can participate independently of aggregators. This is the path that preserves brand identity, margin, and customer relationship in the agentic layer.

Source 3 — Trained knowledge with live verification For some queries, the agent draws on trained knowledge to identify relevant brands — then attempts live verification of availability and pricing. This is where the ESC™ layer becomes critical: if the agent’s trained knowledge includes a clear, accurate, and trusted understanding of a brand’s identity and product range, the brand enters the consideration set even before the live catalogue is checked.

This third source is why building the AI layer — entity clarity, semantic authority, cross-source trust — matters even before the commerce layer is fully connected. A brand that is clearly understood in the agent’s trained knowledge has a presence advantage when live catalogue access is partial or unavailable.

India’s Structural Opportunity — ONDC

The aggregator advantage in inventory legibility exists because aggregator catalogues are connected to agentic platforms by default. For Indian brands, there is a structural counter to this — one that has no equivalent in Western markets.

ONDC — the Open Network for Digital Commerce — is India’s government-backed open commerce protocol. Unlike Amazon’s closed ecosystem, ONDC is a network that any seller can connect to and any buyer-side platform can access. A small saree manufacturer in Varanasi, a D2C skincare brand in Bangalore, a B2B textile supplier in Surat — all can participate on equal technical terms with the largest aggregators.

When AI agents in India begin routing purchase queries through ONDC — which the network’s open architecture makes possible — the inventory legibility advantage currently held by aggregators becomes democratised. Any brand connected to ONDC has the same infrastructure access as any other.

This is not guaranteed. It depends on ONDC’s integration with AI agent platforms, on brands building the structured product data that makes their catalogues legible within the network, and on the agentic platforms choosing to query ONDC as a source.

But the structural possibility exists in India in a way it does not in markets without an equivalent open commerce protocol. This is the ONDC opportunity that most Indian brands are not yet discussing in the context of Agentic Commerce.

The Machine Trust Layer — What Aggregators Cannot Own

There is one dimension of the inventory infrastructure question where aggregators do not have an inherent advantage — and where independent brands can build a position that aggregators cannot replicate.

Machine trust.

Machine trust is not the same as human trust. Human trust is built through brand storytelling, customer experience, social proof, emotional resonance — the things brand marketing has always done. Machine trust is built through data consistency, entity clarity, cross-source verification, and structured signals that AI systems use to evaluate credibility.

An AI agent evaluating a moisturiser on Amazon sees a product listed under a seller account. It does not see a brand. It sees attributes, ratings, and pricing. The brand identity behind the product is flattened by the aggregator layer — the agent knows the product exists and that it meets the stated criteria. It does not know, trust, or recommend the brand itself.

An independent brand that has built machine trust — clear entity declaration, consistent identity across independent sources, verified credentials, structured content that AI systems can read and cite — has something that no aggregator listing can provide: agentic brand recognition. The agent does not just know the product. It knows and trusts the brand.

When a user’s AI agent is instructed to “find me a moisturiser — preferably from a brand I can trust for sensitive skin” — the brand with machine trust enters the consideration set. The brand visible only through an aggregator listing may not.

This is the territory independent brands can build that aggregators cannot own. Not inventory legibility — which aggregators have by default. But machine trust — which brands must build by intention, and which aggregators structurally cannot build for them.

What Agentic Observability Reveals

How does a brand know whether it is currently visible to AI agents — whether through an aggregator or directly?

The emerging answer is agentic observability: the ability to monitor, in real time, how AI agents describe, evaluate, and recommend a brand across the platforms where agents operate.

For most Indian brands, the current state of agentic observability is zero. They have no visibility into whether AI agents are recommending their products, what those agents say about their brand when they do, which sources the agents are drawing from, or why a competitor’s product is being recommended instead of theirs.

Building agentic observability — even at a basic level — starts with a simple test: ask ChatGPT, Perplexity, and Google’s AI to find a product in your category. See what comes back. Note which brands appear. Note whether yours does. Note what the agent says about the brands it recommends and the language it uses to describe them.

This is the beginning of understanding where your brand stands in the agentic layer — and what the gap looks like between where you are and where you need to be.

The Inventory Infrastructure Decision — A Framework

For Indian brands evaluating how to approach the inventory infrastructure question, the decision is not binary — aggregator or direct. It is a sequencing question.

Phase 1 — Establish Machine Trust (now) Build the ESC™ layer regardless of commerce infrastructure:

• Entity clarity — AI systems can identify your brand unambiguously
• Semantic authority — your product information is structured for machine extraction
• Cross-source trust — your identity is consistent and verifiable across independent sources

This phase does not require platform integration. It requires building the AI readiness layer that determines whether your brand enters the agentic consideration set at all.

Phase 2 — Inventory Legibility (now, parallel) Structure your product catalogue for machine readability:

• Attributes declared as structured fields — not buried in description text
• Real-time inventory sync — stock status always current
• Pricing accurate and consistent across all channels

Phase 3 — Commerce Layer Connection (as platforms develop) Connect to the transaction infrastructure:

• Direct API integration with agentic platforms as they become available
• ONDC participation — connect and maintain structured product data
• Payment integration for in-conversation transactions via UPI frameworks

The sequencing discipline: Phase 1 before Phase 3. A brand that connects to the commerce layer before building machine trust is transactable but not discoverable. The agent cannot recommend what it does not know. ESC™ first. Commerce layer second.

The Zero-Click Commerce Reality

The inventory infrastructure question has a destination — zero-click commerce.

Zero-click commerce is the end state of Agentic Commerce for consumer purchases: transactions that complete entirely within the AI interface. No website visit. No cart. No checkout page. No “add to cart” button pressed by a human.

The purchase happens in the conversation. The user’s agent evaluates, selects, and transacts. The user reviews the confirmation.

For brands still investing primarily in website conversion optimisation — checkout flow, cart abandonment recovery, product page UX — this is the implication that changes the investment calculus. The checkout page does not disappear. But for the segment of purchases completed through AI agents, it is bypassed entirely.

The brands that participate in zero-click commerce are the ones that have solved the inventory infrastructure question — legibility, accessibility, and machine trust — before the agent ever reaches the transaction moment.

The brands that have not are simply not available for the transaction. Not because the user chose not to buy from them. Because the agent had no path to do so.