For two decades, ecommerce has been designed around a screen and a human hand. Query, results page, product page, cart, checkout. Every conversion optimisation playbook ever written quietly assumes that a person is looking at something.
That assumption is now coming apart. A growing share of product discovery starts inside an AI assistant rather than a search engine, and a growing share of transactions will finish there too. When the buyer is a model acting on someone's behalf, your homepage hero, your scroll animation and your carefully sequenced upsell flow are invisible. What the agent sees instead is your data.
This is the shift we are building for at NOIR & BLANCO, and this is how we think brands should prepare.
What agentic commerce actually means
The phrase gets used loosely, so it helps to separate three distinct stages. They are arriving in sequence, and most brands are currently exposed to the first one without realising it.
Stage one: AI assisted discovery. A shopper asks an assistant for a recommendation. The assistant reads the open web, retail feeds and its own index, then returns a shortlist. The purchase still happens on your site, but the consideration set was decided before the shopper ever reached you. You did not compete on creative. You competed on how legible your catalogue was to a machine.
Stage two: AI assisted transaction. The assistant carries the shopper all the way to a checkout it controls. OpenAI's Instant Checkout, built on the Agentic Commerce Protocol with Stripe, and Google's Agent Payments Protocol are early expressions of this. The merchant of record is still you. The interface is not.
Stage three: delegated buying. The shopper sets an intent and a budget, and the agent transacts without a final human click. Replenishment, price triggered purchases, gifting within constraints. Card networks have already built the rails for this through Visa Intelligent Commerce and Mastercard Agent Pay, which issue scoped, tokenised credentials to a verified agent rather than handing over a card number.
Stage one is live and material today. Stage two is being wired in right now. Stage three is a question of trust and regulation more than technology.
The uncomfortable part: agents flatten brands
Here is what we keep seeing when we test how assistants describe our clients' products against their competitors.
An agent does not experience your brand. It parses attributes. Material, price, dimensions, delivery window, return policy, warranty, review sentiment. Then it ranks. Everything a premium brand spends money to communicate, the photography, the typography, the pacing of the site, the tactile feel of the product page, does not survive the trip into a model's context window.
That creates a real strategic risk for the category we work in most. If your positioning lives entirely in your art direction, an agent will reduce you to a specification sheet and place you beside a cheaper option with better structured data. The brands that hold their premium in an agentic environment will be the ones that translate craft into claims a machine can carry: verifiable materials, certifications, provenance, warranty terms, care instructions, artisan or production detail written as text rather than baked into images.
Beautiful sites still matter enormously for the humans who arrive. They simply stop being the only surface that sells.
What agents actually read
When an assistant evaluates your product, it draws on a fairly boring set of inputs. In rough order of influence:
Your product feed, and whether it is complete, accurate and syncing. Missing GTINs, empty attribute fields and stale availability are the fastest way to be excluded from a shortlist.
Structured data on your product pages. Schema.org Product, Offer, AggregateRating, availability and shipping details. If your theme renders price in JavaScript and never emits it as markup, you are asking a crawler to guess.
Text on the page. Specifications written as prose, FAQs, sizing guidance, returns language. Anything trapped inside an image or a video is lost.
Third party corroboration. Reviews, editorial mentions, marketplace listings, comparison content. Models weight independent confirmation heavily, which is why review acquisition is now a discovery investment, not just a conversion one.
Machine accessible commerce endpoints. Shopify's catalogue infrastructure and storefront MCP server are quietly becoming the way assistants query inventory and build carts without scraping.
None of this is glamorous work. All of it compounds.
Retail media does not disappear, it relocates
The instinct is to assume agents kill advertising. We think the opposite. Attention concentrates inside a smaller number of assistant surfaces, and monetisation follows attention. Sponsored placement inside AI answers, commerce media inside assistant experiences and paid inclusion in agent shortlists are all being built.
What changes is the unit of competition. Bidding on a keyword assumes a query and a results page. Competing for a slot in an agent's recommendation assumes relevance scoring against a natural language intent that may never repeat in the same form twice. Feed quality, product data richness and post purchase signals such as return rate become bidding inputs, not just operational hygiene.
For our paid media clients, the practical implication is immediate: the same feed that powers Performance Max and Advantage+ catalogue campaigns is the asset that determines agent visibility. Fixing it pays twice.
Measurement is about to get harder before it gets easier
Assistant driven traffic arrives with thin or missing referrer data, and a meaningful share of influence produces no click at all. A shopper reads a recommendation, then types your brand name into a browser two days later. Your analytics call that direct.
Three things we are putting in place for clients now:
Custom channel groupings and regex based referral rules in GA4 to isolate known assistant domains, imperfect but far better than nothing.
Server side tracking so that conversion signal survives when the browser context is stripped.
A post purchase survey question asking how the customer first heard about the brand, which is currently the single most honest measurement instrument available for zero click influence.
Expect branded search volume and direct traffic to become proxy indicators of AI visibility. Watch their trend line, not their absolute value.
The India layer that global commentary keeps missing
Most writing on agentic commerce assumes a card on file and a single click. The Indian market is structured differently, and it changes the sequencing.
Cash on delivery. If an agent places an order and the human never explicitly confirmed it, return to origin risk rises sharply. Prepaid conversion, risk scoring at checkout and address quality tooling become prerequisites for safe agentic buying, not optional upgrades. Our clients running GoKwik and similar stacks are better positioned here than they realise.
WhatsApp as the agent surface. For a large part of the market, the assistant will not be a browser tab. It will be a conversation thread with catalogue, payment and support in one place. Brands with clean catalogue sync into that channel have a head start.
ONDC. Whatever one thinks of adoption to date, a protocol that separates buyer applications from seller applications is structurally an agentic substrate. It is worth watching as an interoperability layer rather than as a marketplace.
Marketplace gravity. Amazon and Flipkart hold enormous transaction data and are building their own assistants. Owned site optimisation does not remove the need to be legible and well rated where the volume already sits.
The NOIR & BLANCO playbook
This is the work we are doing with brands right now, in the order we do it.
1. Data foundation. Full catalogue audit. Attribute completeness, identifier coverage, variant hygiene, structured data emitted server side on every product page, feed diagnostics across every channel. This is where most brands lose before they start.
2. Content that machines can carry. Rewriting product content so that materials, dimensions, care, provenance and sizing exist as text. Building FAQ and comparison content that answers the questions people actually ask assistants, in the language they ask them.
3. Corroboration. Review volume and recency programmes, editorial and creator placement chosen for indexability rather than reach alone, and consistency of brand facts across every surface a model might read.
4. Agent ready commerce. Checkout and inventory that can respond to programmatic requests. Clean APIs, accurate real time stock, transparent shipping and returns policies expressed as data rather than as a PDF.
5. Measurement and iteration. Assistant visibility testing on a recurring cadence, tracking how each brand is described and ranked against named competitors, with the gaps fed back into steps one through three.
A ninety day starting point
Days 1 to 30. Catalogue and feed audit. Structured data implementation across product and collection templates. Baseline assistant visibility test across the major assistants for your top twenty commercial queries.
Days 31 to 60. Product content rewrite for the top revenue SKUs. Review acquisition programme live. Server side tracking and AI referral channel grouping in place. Post purchase attribution survey deployed.
Days 61 to 90. Comparison and FAQ content published. Feed extended with enriched attributes into paid channels. Second visibility test to measure movement. Prepaid conversion and address quality work for the Indian checkout risk profile.
Where we think this lands
Agentic commerce will not replace brand. It will separate brands that have substance from brands that have only styling. A model cannot be seduced, but it can be convinced, and what convinces it is verifiable, well structured, independently corroborated fact.
The brands that win the next five years will be the ones that made their products legible to machines without making them boring to people. That is a design problem and a data problem at the same time, which is precisely the intersection we work in.
If you want to know how your brand currently reads to an AI assistant, we can show you. It is usually the most uncomfortable and most useful hour a founder spends this quarter.
NOIR & BLANCO builds and grows ecommerce brands. Shopify design and development, paid media, and the data infrastructure underneath both.
Top comments (0)