Is agentic commerce a board-level decision or an e-commerce team decision?

It is a board-level decision with an e-commerce implementation. The strategic question — whether to invest now in being purchasable by AI agents, or to defer and accept the discovery risk — is a capital-allocation and risk call that belongs with the board and the CAIO, not buried in a digital-team roadmap. The implementation (product feeds, real-time data, an agent-callable checkout) is operational. What makes it board-level is that the downside of doing nothing is silent: you do not get an outage alert when an agent skips your catalog because it cannot parse you. The decision to be invisible to the agent channel is usually made by default, which is exactly why it should be made deliberately.

What is the actual investment to become agent-ready?

Smaller than a replatform, larger than a config change. The four building blocks are a clean product feed with stable identifiers, real-time price and stock served programmatically, machine-readable shipping and returns and tax policies, and a checkout an agent can call through a commerce protocol wired to a consent-bound payments layer. Organizations with a mature data foundation and a modern commerce platform are close; those carrying product-data debt will find that debt is the real cost. The right framing for the board is that most of this spend is data-quality and API work you can justify on its own merits, with agent-readiness as the forcing function rather than the sole rationale.

Which protocols should we standardize on — ACP, AP2, or the card networks?

They are complementary layers, so the question is sequencing, not selection. The Agentic Commerce Protocol (OpenAI and Stripe) governs the checkout interaction; the Agent Payments Protocol (Google, now donated to the FIDO Alliance) carries the cryptographic proof of customer consent; Visa's Intelligent Commerce Connect and Mastercard's Agent Pay provide acceptance on existing card rails. Critically, the card networks shipped protocol-agnostic on-ramps in 2026, which means you can accept agents across several standards through one integration and avoid betting on a single protocol winning. The board guidance is to integrate the commerce layer first for discoverability, wire payments through your existing processor's agentic support, and treat the standards landscape as still-consolidating — short contracts, no lock-in.

What is the cost of waiting?

Asymmetric and quiet. The cost of preparing is a bounded, justifiable data-and-API investment. The cost of waiting is losing the routine, high-frequency purchases — reorders, replenishment, spec-matched buys — to competitors who are parseable when an agent goes looking, and you will not see those losses in any dashboard because the agent simply never reached your checkout. The honest board framing is not 'agentic commerce will be huge, invest now,' which is unprovable, but 'being unbuyable by agents is a position we should choose on purpose if we choose it at all.' That reframes a speculative bet into a risk-management decision.

What governance questions does agentic commerce raise?

Three the board should put on the table. First, consent and liability: when an agent transacts, the payments-layer mandate is what proves the customer authorized the spend — your processor and legal team need to be clear on where liability sits for a disputed agent purchase. Second, pricing exposure: agents are efficient price-discoverers comparing you against alternatives in milliseconds, which raises the stakes on pricing governance and surfaces any inconsistency faster. Third, data and trust: exposing structured product, inventory, and policy data to agents is a new external surface that should run through the same governance as any other data-sharing decision. None of these are blockers; all of them are questions that are cheaper to answer before launch than after.

Agentic Commerce Strategy: What the Board Needs to Decide Before AI Agents Buy

Tom Prommer · CIO/CTOUpdated 2026-06-1012 min read

Most strategic technology shifts announce themselves with an outage, a competitor launch, or a line item that suddenly doubled. Agentic commerce will not. The first sign that it matters to your business is an absence you cannot measure: the routine purchases an AI agent made somewhere else because it could not parse or buy from you. That is what makes it a genuine board question rather than a digital-team roadmap item — the downside of inaction is silent, and silent risks are the ones organizations get wrong.

What agentic commerce actually is

Agentic commerce is buying done by an AI agent on a customer’s behalf. The customer sets the intent — reorder what fits my equipment, find an option under this budget that ships by Friday — and an agent inside an assistant like ChatGPT, an answer engine like Perplexity, or a retailer’s own copilot does the discovering, deciding, and checking out. The defining capability is that the agent transacts, not merely recommends. It reaches the merchant through an open commerce protocol and completes payment with a cryptographic mandate proving the customer authorized that exact spend.

This is distinct from the chatbots most enterprises already run. A chatbot suggests and hands the customer to a website. An agent completes the purchase. That single difference relocates the competitive surface: the things you optimize for a human shopper — imagery, reviews, persuasive copy, a frictionless cart — matter less to the agent, and the things you may have under-invested in — clean structured data, real-time signals, a callable checkout — become the determinant of whether the agent can buy from you at all.

The strategic decision, stated honestly

It is tempting to frame this as “agentic commerce will be enormous, so invest now.” That framing is unprovable and invites the board to discount it as hype. The honest framing is the inverse: being unbuyable by agents is a position your organization will hold by default unless you decide otherwise, and a default is not a decision. The protocols and the card-network rails both shipped in early 2026, moving agentic commerce from concept to live channel. The question is no longer whether the rails exist. It is whether you want to be reachable on them, and that is a deliberate choice the board should make rather than back into.

The asymmetry is what makes the call straightforward. The cost of preparing is a bounded, independently justifiable investment in data quality and API surface. The cost of waiting is losing high-frequency, low-deliberation purchases — the reorders and replenishment that make up a large share of commerce — to whoever is parseable when an agent goes looking. You will not see those losses; they never reach your funnel.

What “agent-ready” requires

The readiness work is smaller than a replatform and larger than a configuration change. Four building blocks carry it:

A clean product feed with stable identifiers and machine-readable attributes. Ambiguous or incomplete product data is the most common reason an agent skips a merchant.
Real-time price and stock, served programmatically rather than only rendered on a page, because the agent commits to a purchase based on the data it reads at decision time.
Machine-readable policies for shipping, returns, and tax, so the agent can reason about total cost and eligibility before it transacts.
An agent-callable checkout, adopting a commerce protocol and wired to a payments layer that clears only with provable consent.

Organizations with a disciplined data foundation and a modern commerce platform are closer to this than they think. Those carrying years of product-data debt will discover that the debt is the real cost — which is also the strongest argument for the investment, because clean product data and real-time inventory APIs pay for themselves across search, marketplaces, and analytics regardless of how large the agent channel becomes. Agent-readiness is best sold to the board as the forcing function for data-quality work you could justify anyway, not as a speculative bet standing on its own.

The protocol stack, without the format-war framing

The standards landed quickly and the coverage makes them sound like competitors fighting for dominance. They are layers doing different jobs:

ACP — the Agentic Commerce Protocol (OpenAI and Stripe) governs the checkout interaction: how an agent discovers what it needs and submits an order.
AP2 — the Agent Payments Protocol (originated by Google, now donated to the FIDO Alliance for neutral governance) carries the cryptographic mandate proving the customer authorized the spend.
Card-network frameworks — Visa’s Intelligent Commerce Connect and Mastercard’s Agent Pay — provide acceptance on existing rails, and shipped protocol-agnostic on-ramps that accept several agent standards at once.

For the board, the implication is reassuring: the protocol-agnostic acceptance layer means you do not have to predict which standard wins. Integrate the commerce layer first for discoverability, wire payments through your existing processor’s agentic support, and treat the landscape as still consolidating — short contracts, no lock-in, an explicit eighteen-to-twenty-four-month review horizon.

The governance questions to put on the table

Three questions belong in front of the board before any launch, and all three are cheaper to answer now than after an incident.

Consent and liability. When an agent transacts, the mandate is what proves authorization. Your processor and legal team need a clear position on where liability sits for a disputed agent purchase.
Pricing exposure. Agents are efficient price-discoverers, comparing you against alternatives in milliseconds. That raises the stakes on pricing governance and surfaces inconsistency faster than human shoppers ever did.
Data and trust. Exposing structured product, inventory, and policy data to external agents is a new data-sharing surface and should run through the same governance as any other.

None of these are blockers. They are the questions that distinguish a deliberate entry into the channel from an accidental one.

The recommendation

Make the decision on purpose. Authorize a bounded, sequenced investment to make your highest-velocity, most-reordered products parseable and buyable by agents, justified primarily on the data-quality merits and secondarily on channel readiness. Integrate the commerce layer first, wire provable-consent payments through your existing processor, and answer the three governance questions before launch. Hold the standards landscape at arm’s length with short contracts. And name the alternative for what it is: choosing to be invisible to the agent channel is a strategy too — just make sure it is the one you meant to choose.