The pipe exists. The door is locked. That was the Wall Street Journal's verdict on agentic commerce last week, and it remains the most precise summary of where this market actually stands. But something shifted on October 5. Constructor, a retail AI search and discovery firm, integrated with Stripe to let AI shopping agents complete purchases directly inside a conversation. Not route the user to a checkout page. Not generate a cart link. Execute the transaction. For the first time, the 'welcoming' retail posture — the one QVC adopted, the one Shopify hardcoded as default, the one Amazon rejected outright — has a scalable execution endpoint behind it.
Understand what this is not. This is not a product launch. This is the structural moment when discovery and transaction collapse into the same surface. The browse-to-buy gap — the friction layer that e-commerce optimization firms have been monetizing for two decades — moves inside the AI agent. That is a profound shift in where value accrues. It is also where the responsibility gap becomes operational rather than theoretical.
I audited smart contracts for three months in 2018. Seven reentrancy vulnerabilities in 0x protocol v2. The lesson I carried into every trade since: code is law, but liquidity is truth. The same logic applies here. Constructor's code — or rather, its Commerce Reasoning Engine, which orchestrates product discovery, recommendation, and natural language resolution — can now route a purchase intent to a Stripe payment endpoint. The code path is closed. The economic path — who pays when the agent gets it wrong — remains wide open. And in a bear market, that gap is where capital goes to die.
The Chain That Now Exists
Map the actual structural layers. This matters because most commentary treats 'agentic commerce' as a monolith. It isn't.
Application layer: Constructor's Commerce Reasoning Engine, AI Shopping Agent, and AI Product Insights Agent handle discovery. They parse natural language queries. They recommend. They resolve 'I need a black dress for a wedding under $300' into a product set.
Transaction layer: Link by Stripe completes payment in-conversation. Merchants already on Stripe need only a configuration change. Merchants on other processors can route through Constructor's own Stripe integration without switching their primary payment processor. This is the critical adoption detail. The switching cost is near-zero for the merchant's tech stack.
Network and standards layer: Visa's TAP and Mastercard's Agent Pay provide signals that identify agent-initiated transactions. Amex announced Agent Purchase Protection in April 2026. The Cart Context technical specification — which would define how purchase intent is verified — remains under development. Read that again. The specification for verifying whether an agent's purchase matches the human's actual intent does not exist yet. It is being drafted. Meanwhile, live transactions are processing.
Merchant and platform layer: Sephora, Under Armour, The Very Group, Monica Vinader, White Stuff, Furniture Village, Target Australia, AMI Paris are live. The Wall Street Journal documented three distinct postures: QVC welcomes agents, Kate Spade allows browsing but blocks purchase, Amazon blocks entirely. Shopify hardcoded agent-readiness as the default for its merchant base. When Shopify makes something default, adoption becomes the path of least resistance. That is a structural forcing function.
The real significance: the transaction endpoint moved from the browser interface into the AI agent's reasoning loop. The merchant no longer owns the final conversion surface. The agent does. And the agent — whether Constructor's or another provider's — optimizes for the human's stated intent, not the merchant's margin structure.
Four Tensions That Will Determine Survival
The Posture-Infrastructure Mismatch
QVC saying 'we welcome agents' cost QVC nothing. There was no pipe for the agent to buy through. It was a press-release posture. Constructor's integration changes that. 'Welcoming' now means providing a live purchase endpoint. The low activation cost is the accelerant: a configuration change, not a re-platforming. For Stripe merchants, the barrier drops to near zero. For non-Stripe merchants, Constructor routes through its own Stripe integration — the merchant keeps their primary processor. This is brilliant adoption engineering. It also means merchants can walk into agentic commerce without building the operational infrastructure to handle it.
The Readiness Gap
The numbers are brutal. Only 11% of small and mid-sized merchants report being 'agent-ready' in any operational sense. Only 15% have structured product data that an AI agent can parse reliably. Only 23% can distinguish AI traffic from human traffic in their analytics. These are not minor gaps. They are the difference between an agent completing a correct purchase and an agent completing a wrong one.
Structured product data is not a nice-to-have. When an agent resolves 'black dress, wedding, under $300' into a SKU, it is querying structured attributes — color, occasion, price, size availability, material. If the merchant's catalog is PDFs and unstructured descriptions, the agent guesses. Guesses convert to returns. Returns convert to margin destruction.
AI traffic identification matters for a different reason: without it, the merchant cannot measure agent-driven conversion separately from human-driven conversion. They cannot isolate the error rate. They cannot price the risk. Running agentic checkout blind is not a strategy. It is exposure.
Shopify's default-on policy drove an 8x increase in AI-driven traffic in Q1 2026, according to the data. Eight times more agent sessions. Eight times more opportunities for the readiness gap to convert into operational loss. The merchant's onboarding experience is one configuration change. The operational reality is months of data infrastructure work that most merchants have not done.
The Liability Void — This Is the Real Story
Here is the number that should stop every merchant executive cold: 93% of merchants believe the AI provider should bear the financial loss when an agent makes a wrong purchase. In the current model, the merchant bears 100% of it.
The agent selects the wrong size. The merchant eats the return. The agent misinterprets the color preference. The merchant eats the exchange. The agent processes a purchase the human never confirmed. The merchant eats the chargeback. Visa TAP and Mastercard Agent Pay identify that an agent initiated the transaction. They do not reallocate responsibility for non-fraudulent agent error. There is no reason code for 'agent error.' There is no dispute category for 'agent exceeded buyer intent.' The Amex Agent Purchase Protection program is a signal that the card networks recognize the gap. It is not yet a solution. The Cart Context specification, which would define purchase intent verification, is still being drafted.
The payment networks can identify that an agent made a purchase. They cannot, or will not, identify whether that purchase matched what the human actually wanted. Until that gap closes, the merchant is the insurer of last resort for every AI mistake.
And here is the part that keeps me up at night as a former options strategist: this is an uncapped liability. Every agent transaction is a potential dispute. The merchant cannot price it. They cannot hedge it. They cannot reserve against it accurately because there is no loss history. They can only absorb it and hope the conversion lift exceeds the error cost. In a bear market, where margins compress and every basis point of operational efficiency matters, that is an unacceptable risk profile. It is also the only risk profile currently available.
The Unit Economics Question
Enabling agentic checkout is a bet. Specifically, it is a bet that Constructor's reasoning engine — or whichever agent provider the merchant integrates — will execute purchases that align with the customer's actual intent. If the bet wins, the merchant captures conversion volume they would have lost to friction. If the bet loses, they absorb return processing, dispute resolution, chargeback fees, and the customer's frustration.
The technical infrastructure is ready. Stripe's payment rails are the most battle-tested in the industry. Constructor's reasoning layer handles natural language product resolution. The code path executes. But nobody has published the aggregate return rate for agent-initiated purchases versus human-initiated purchases. Nobody has published the dispute rate. Nobody has published the chargeback rate. The unit economics are unverified. The merchant is flying blind into a live transaction environment.
Contrarian: The Liability Gap Is the Product
Every analyst covering this integration is focused on the adoption question: will merchants enable agentic checkout? That is the wrong question. The right question is: what is the actual product being sold here?
Stripe and Constructor are not selling payment processing. They are selling conversion volume that friction currently kills. The pitch is straightforward: your human customers abandon carts at 70% because the checkout process is too long, too complicated, or interrupted. An AI agent that completes the purchase in-conversation recovers that volume. The merchant pays a fee on recovered volume they would never have seen.
That is the surface pitch. The deeper structure is different. By enabling agentic checkout, the merchant is not just recovering abandoned carts. They are accepting liability for an automated system's errors with no pricing mechanism for that liability. The 93% of merchants who believe the AI provider should bear the loss are correct in theory. In practice, they are signing contracts that transfer that liability to themselves without a corresponding price adjustment.
This is not a conspiracy. It is the natural equilibrium of a market where the infrastructure provider has more power than the merchant. Stripe controls the payment rails. Constructor controls the reasoning layer. The merchant controls the product catalog and the fulfillment operation. When the transaction completes and the agent is wrong, the merchant cannot return the product to Stripe. They cannot file a claim with Constructor. They absorb the loss and adjust future behavior. The only leverage the merchant has is the threat to disable agentic checkout. In a bear market where every competitor is enabling it, that is not leverage. That is a race to the bottom of the risk curve.
Watch the contract terms. Watch whether merchants negotiate liability caps or indemnification clauses with Constructor and Stripe. Watch whether the card networks move from identification to reallocation. Watch whether the Cart Context specification includes intent verification as a mandatory field. Those three signals will tell you whether the liability gap is a temporary technical problem or a permanent structural feature of agentic commerce.
What to Watch — Forward-Looking Indicators
The Cart Context specification is the single most important document in agentic commerce right now. If it defines purchase intent verification and maps it to liability allocation, the merchant's exposure shrinks. If it only defines data formats for passing context between agent and checkout endpoint, the liability gap persists. Watch the draft releases. Watch who is participating in the working group. Watch whether the liability language is mandatory or optional.
The card networks' next moves matter. Visa TAP and Mastercard Agent Pay currently provide identification signals. The question is whether they will extend into non-fraudulent error liability. If Amex's Agent Purchase Protection moves from a PR program to a contractual guarantee with published coverage limits and claim conditions, it becomes a pricing mechanism. Other networks will follow. If it remains a marketing program, the liability void persists.
Watch the merchant error rates. The first public data on agent-driven return rates, dispute rates, and chargeback rates will reset the entire conversation. If the rates are comparable to human-initiated transactions, the liability concern is overstated. If they are materially higher — and in a system where the intent verification spec does not exist, they will be — merchants will renegotiate contract terms or disable the feature. The data speaks louder than sentiment. Right now, there is no data.
Watch the structured data investment cycle. Only 15% of merchants have agent-ready product data. The merchants who invest in structured catalogs, AI traffic analytics, and agent-specific operational workflows will capture the conversion volume. The merchants who do not will absorb the errors without the upside. In a bear market, the capital discipline required to make that investment separates survivors from casualties. The survivors will have done the data work before the volume arrives. The casualties will have chased the conversion without building the infrastructure to handle it.
The pipe exists. The door is now unlocked. But behind the door is a liability void that no one owns yet. The merchants who walk through it without a contractually defined risk framework are not early adopters. They are unpaid beta testers for the card networks' next revenue line.
