How Akwa Is Tackling Agentic Commerce
By Egoyibo Okoro · August 2026
For most of ecommerce history, fashion has been designed for a human moving through a website.
You search. You filter. You open a product page. You choose a size. You add to cart. You check out.
Agentic commerce changes that sequence.
Increasingly, the customer may never begin on the brand's website. They may tell an AI agent:
Find me a structured pair of wide-leg trousers in natural fibres, suitable for a wedding in Lagos, under 400 euro, that I can wear with the ivory jacket I already own.
The agent has to understand that request, discover suitable products, compare them, determine whether they are actually available, understand fit and delivery constraints, and potentially complete the transaction.
That creates a very different design problem for fashion technology. At Akwa, we are preparing for it by asking a simple question: what does a fashion product need to look like to a machine?
A product page is no longer enough
Traditional ecommerce has spent decades optimising information for human eyes. Beautiful photography. Editorial copy. Navigation. Filters. Product descriptions.
Agents need something else as well: structured, reliable and machine-readable product truth.
Stripe's Agentic Commerce Protocol allows compatible applications and sellers to interact through REST APIs or MCP servers, while the seller retains its underlying data model and payment processing. Google is moving in a similar direction with the Universal Commerce Protocol, an open standard co-developed with Shopify, Etsy, Wayfair, Target and Walmart, built on REST and JSON-RPC with support for the Agent Payments Protocol, Agent2Agent and the Model Context Protocol. Stripe is among its endorsers.
Two protocols, converging on the same requirement. For Akwa, the answer is not to build a different commerce system for every agent. It is to build one governed commerce truth that can be projected into many channels.
Three layers, and they must not collapse
Akwa's architecture separates three things that are easy to confuse.
The Semantic Garment Graph is our authoritative fashion layer. It represents what Akwa knows about a design: construction, silhouette, materials, provenance, cultural context, design intent, and the relationships between fashion concepts. Crucially, it also holds where each of those assertions came from.
The Canonical Commerce Product Object does not replace it. Its job is to operationalise the semantics for commerce and combine them with transactional state that has no business living in an ontology: price, currency, sizes, availability, inventory, preorder status, destination eligibility, shipping, duties, checkout state.
Channel and protocol feeds are projections of that object. Stripe is one projection. Others should consume the same governed truth rather than maintain independent interpretations of the same garment.
The discipline in that separation matters more than the diagram. The commerce object must never become a second ontology competing with the graph, because two descriptions of the same garment will eventually disagree, and the customer will meet whichever one is wrong.
Fashion cannot be represented by a SKU and a photograph
A customer might care that a pair of trousers is wide-legged. They might also care that the fabric has enough structure to hold its silhouette, that it is made from a natural fibre, that the outfit suits a particular occasion, or that its cultural references are appropriate to the context in which they intend to wear it.
So alongside the commercial facts, Akwa's commerce object carries a semantic layer: silhouette, material and weight, construction, styling relationships, design notes, cultural register and design provenance.
We are not merely trying to make Akwa's products searchable by agents. We want to make them understandable by agents.
Publishing what we know, and publishing what we do not
There is a governance rule underneath that layer which we consider more important than the layer itself.
Every attribute Akwa publishes carries its source. Not a confidence score, a source: the ratified copy, the founder decision or the measured fact it was read from. Where Akwa has not established an attribute, we publish that gap rather than allowing a downstream system or an agent to invent an answer.
Fit is the clearest example. It is the attribute an agent most wants when a customer says they need an XL, and it is the one most likely to cause a return if it is guessed. Akwa currently publishes fit as unstated for Drop 01, because the measurements that would make it true are still being established. An agent reading our catalogue can say so honestly. An agent guessing creates a return, a refund and a disappointed customer.
The question is never whether something sounds plausible. It is what does Akwa actually know about this design, with sufficient provenance to assert it.
Culture is not an optional field
That same rule produced a correction worth describing, because we nearly got it wrong.
When we first modelled cultural context, a denim collection appeared to have none to state. Drop 01 is denim. There is no visible heritage motif to point at.
That reading was wrong, and it contradicted our own thesis. Akwa's position is that heritage is structural DNA, not a styling overlay. A tuxedo has cultural provenance. Jacquard has aesthetic lineage. Denim carries context as surely as akwete does. The governance distinction is not between cultural clothes and plain ones. It is between what a house can evidence and what it cannot.
So Drop 01 publishes its cultural register explicitly: origin-native design, expressed through cut and construction rather than print, with Nigerian heritage present as one register among many rather than a default. A house that tells a machine its own denim has no cultural context has conceded its argument before the conversation starts.
Inventory becomes part of the conversation
An agent should not recommend a product because a static catalogue says it exists when the customer's size has sold out.
For limited editions the distinction matters more. If fifty pieces exist, the machine-readable representation should understand scarcity as a live commerce fact rather than marketing language pasted onto a product page.
That is why Akwa separates its product representation from its inventory representation. The product describes what the thing is. Inventory describes what can actually be bought now. When an order is recorded, the inventory projection updates within seconds rather than waiting for a scheduled refresh, because a count that is stale by hours is a count that sells something twice.
The same principle applies to preorders. A product that becomes available on a future date should not masquerade as ordinary in-stock inventory. That distinction allows an agent to reason honestly about the purchase it is proposing.
Discovery before autonomous checkout
We are deliberately separating agentic discovery from agentic checkout.
The exciting demonstration is an AI agent buying something. The difficult work begins when that transaction touches a real commerce system.
What happens when only three units remain? What if the product is a preorder rather than immediately available? Which price does the agent receive if a promotion is active? What happens to attribution? Does the transaction decrement the same inventory as a website purchase? Does the order enter the same fulfilment workflow? What duties information must the customer see before authorising payment? What happens if the payment succeeds but the order-recording webhook fails?
These are not AI questions. They are commerce infrastructure questions, and we have answered them for our own checkout the hard way, which is the only reason we know to ask them of somebody else's.
So Akwa's products become increasingly legible and discoverable to agents while agent-initiated checkout stays behind a gate until the complete payment and order lifecycle has been validated against the same controls our direct commerce already uses. That gate is a single governed flag rather than a setting buried in a feed, so enabling it is a decision rather than an accident.
Agentic does not have to mean uncontrolled.
What exists today, and what does not
We think it is worth being precise about this, because architecture that makes something possible is not the same as having built it.
Today, Akwa projects its canonical commerce object into its own machine-readable interface and into Stripe-compatible product and inventory feeds. Those feeds are live, and the catalogue is discoverable.
The architecture is deliberately protocol-independent. As agentic commerce standards mature, the same governed object can be projected into additional protocols rather than rebuilding Akwa's product truth for each channel. Those adapters do not exist yet. Neither does agent-initiated checkout. We would rather say so than describe a roadmap in the present tense.
The Semantic Garment Graph is live substrate for design, and connecting it as the upstream source for commerce semantics is the next piece of work rather than a finished one. Commerce has given that graph an economic job it did not previously have, which is a better reason to finish it than any we had before.
Fashion discovery can become much more interesting
This is where agentic commerce becomes genuinely exciting for fashion.
Today, ecommerce discovery is still largely taxonomy driven. Women, clothing, trousers, wide leg, beige.
People do not think about clothes that way. They think: I need something for my cousin's traditional wedding. I want trousers that feel dramatic without being uncomfortable. Find something that works with the jacket I bought last month. I only wear natural fibres. I like this silhouette, but I want something that references my own cultural context.
An agent capable of understanding the customer and querying sufficiently rich fashion data can move beyond filters towards intent-based discovery. And if that agent can one day interact with Akwa's Digital Twin, with the customer's permission, the question stops being what does Akwa sell, and becomes what does Akwa sell that makes sense for this person.
The transaction should still belong to the customer
There is a governance question hiding inside all of this.
An agent may discover the product. Another service may provide the conversational interface. A commerce protocol may coordinate checkout. A payment provider may process the payment. Behind all those systems is still a person making a purchase.
Our approach therefore follows the same philosophy we apply elsewhere in the platform: automation should increase capability without making accountability disappear.
The customer should know what is being purchased, what it costs, when it will arrive and what conditions apply. The merchant should be able to trace how the transaction occurred. The agent should receive authoritative product information rather than inventing missing facts. And the system should preserve provenance as information moves between participants.
Trust is not something to bolt onto agentic commerce after it scales. It belongs in the architecture.
From websites for humans to commerce infrastructure for both
Websites are not disappearing. People will still want to browse collections, see campaigns, understand designers and sometimes simply enjoy fashion.
But the website is becoming one interface to the commerce system rather than the commerce system itself.
A customer may discover Egoyibo Okoro through an editorial photograph. Another may arrive through search. Another through social. And another may simply tell an assistant what she needs and never consciously visit Akwa at all.
The underlying product truth should survive every one of those journeys.
That is what we are building towards: a fashion system that can speak to people, platforms and agents without losing the meaning of the clothes somewhere in between.
Agentic commerce is therefore not a checkout feature for Akwa. It is another expression of something we have been building from the beginning: fashion as structured, governed and interoperable intelligence.
And if the next generation of customers increasingly arrives through agents, we want Akwa to be ready to meet them there.