What is agentic commerce readiness?
Most explainers answer "what is agentic commerce" for a consumer or an analyst. This one answers a narrower question: what it means for your store and the people who sign off on your revenue. Agentic commerce readiness is the practical state of being found by, sold to, and controlled around AI shopping agents. It rests on three pillars, and a store can pass one while failing the other two: get found, get bought, and stay in control.
How do AI shopping agents find and buy from your store?
An agent does not browse the way a person does. It queries a structured catalog, filters against a shopper's constraints, confirms live price and stock, then attempts to complete a purchase programmatically. The full merchant path is three steps that no incumbent explains end to end: get found (your catalog is machine-readable and current), get bought (an agent can complete checkout without a human keystroke), and stay in control (your finance and risk team can prove every agent-driven order was authorized and bounded). Miss any step and the sale either never reaches you or reaches you in a form your controller will not accept.
What does agentic commerce look like in practice?
A shopper tells an AI assistant: "Find trail-running shoes under 140 dollars, size 44, in stock, that ship to Ireland this week." For a mid-market outdoor retailer, the agent reads the structured catalog, matches variant, GTIN, price and stock, checks the shipping policy in machine-readable form, and completes checkout with a scoped payment credential. The retailer never sees a session, a click or a cart. This is not hypothetical demand. Adobe reported that AI-sourced traffic to US retail sites converted 42 percent better than other traffic in March 2026, a full reversal from a year earlier, and warned that large portions of retail sites are still not machine-readable enough for agents to use (Adobe). McKinsey estimates agentic commerce could orchestrate 3 trillion to 5 trillion dollars globally by 2030, with up to 1 trillion in the US alone (McKinsey).
How is agent readiness different from SEO and conversion optimization?
SEO optimizes for human clicks and rankings. Conversion optimization tunes the human checkout experience. Agent readiness optimizes for a machine consumer whose success metric is a completed, bounded transaction, not a page view. A store can rank first on Google, convert humans beautifully, and still be invisible and unbuyable to an agent because its catalog is not parseable, its checkout needs a human, and its finance team has no record it can defend. Readiness is a different discipline for a different buyer.
Get found: can AI agents discover your products?
This is pillar one. Discoverability for agents is not the same as being visible in an AI answer. It is whether the structured catalog behind your store is complete, current and correct when an agent queries it.
What makes a product feed readable by AI agents?
A readable feed is the structured catalog an agent actually queries: every variant, GTIN, attribute, price and stock level, exposed with real-time accuracy, plus machine-readable policies for shipping, returns and eligibility. An agent will not guess. If a size, a ship-to country or a returns window is not in the structured data, the product is filtered out of the shopper's result set before you ever compete on it. We cover this layer in depth in the product feed agents can read.
Do you need structured data, JSON-LD and llms.txt?
For the discoverability layer, yes. Schema.org Product markup in JSON-LD gives agents a typed, unambiguous read of each product. An llms.txt file gives agents a map of what your site offers and how to consume it. Neither is decorative: they are the difference between an agent parsing your catalog and skipping it.
What are the 12 signs your product data is invisible to AI agents?
The failure modes cluster into three groups: stale or missing attributes (no GTIN, no variant-level stock, a price that lags reality), absent structured markup (no JSON-LD, no machine-readable policy), and a catalog exposed only through a human interface with no agent-readable endpoint. If any of these describe your store, agents are silently dropping your products. We break down the full set in 12 signs your product data is invisible to AI agents.
Get bought: can an agent complete the purchase?
Pillar two is the warmest near-term money question. Being found is worthless if the agent hits a wall at checkout.
What is an agent-ready checkout?
An agent-ready checkout is one an authenticated agent can complete end to end, programmatically, with a scoped payment credential and a machine-readable confirmation, without a single human keystroke. If your checkout needs a person to click, type a card, or solve a challenge, it is not agent-ready, no matter how good it is for humans.
Do you need ACP, MCP or AP2 to be agent-buyable?
You do not have to pick a winner, and you should not bet the store on one name. The layers do different jobs. The Model Context Protocol, Anthropic's open standard, is the read and tool layer that exposes your catalog and actions to an agent. The Agentic Commerce Protocol, codeveloped by OpenAI and Stripe, standardizes the buy and checkout step for ChatGPT and Copilot (Stripe); ChatGPT's in-chat Instant Checkout was retired in March 2026, and ChatGPT now finds products through merchant feeds. The Universal Commerce Protocol, from Google with Shopify, lets agents in Google AI Mode, Gemini and Copilot discover a store's capabilities at /.well-known/ucp and check out. Google's Agent Payments Protocol addresses payment mandates. The order that matters: implement the read layer that lets agents discover and evaluate your catalog, then the agent checkout. We stay protocol-agnostic and wire the layers your channels actually use.
How does an MCP server expose your catalog to shopping agents?
An MCP server presents your catalog as tools an agent can call: search products, get live price and stock, read a policy, initiate a checkout. The agent reasons; the server enforces what is real. This is exactly the layer Saturnia Labs runs as a thin MCP server over your existing store API, aligned with the ACP checkout standard, so what we describe is what we operate rather than a diagram on a slide.
Stay in control: can your finance team sign off on agent-driven sales?
This is the pillar every free scanner and checklist skips, and it is the whole differentiator. Discoverability and checkout are being commoditized by free tools. The governance content that exists is buyer-side: how a company stops its own agents overspending. This is the other side of the transaction: how you, the merchant, stay in control of the orders agents place with you, and can prove it after the fact.
How do agent spend limits and approval workflows work?
Controls are set below the model, not asked of it. That means an enforced spend cap (per agent, per transaction, per period), allowlists of which agents and payment methods may transact, and an approval gate: auto-approve under a defined threshold, step up to human approval above it. The design principle the industry now repeats is that prompts are instructions, not controls. A limit only counts if the checkout and order layer enforces it regardless of what the agent was told.
What is an AI agent audit trail?
An AI agent audit trail is a tamper-evident record of every agent-driven order: which agent acted, under what signed mandate and authorization, what it executed, the price and terms at the moment of purchase, and the confirmation returned. This is the record a controller reconciles, an auditor accepts, and you use to defend a chargeback. Without it, an agent order is an unexplained line item.
Is agentic checkout safe and compliant for merchants?
It is when control is enforced at the order layer and mapped to finance sign-off. This is not optional polish. Gartner predicts that more than 40 percent of agentic AI projects will be canceled by the end of 2027, citing inadequate risk controls among the causes (Gartner). The stores that keep agent revenue are the ones whose risk teams could sign off before the first agent order, not after.
How is agentic commerce readiness scored?
Free scanners return a number and a generic gap list as lead generation. Some advertise "31 checks" or "147 checks" with no published method. A score you cannot inspect is not evidence. A useful rubric is protocol-agnostic, inspectable and reproducible: each criterion scored 0 (absent), 1 (partial), 2 (present) or 3 (verified in a live agent transaction), then weighted to a 0 to 100 composite that measures all three pillars in one pass and treats governance as a first-class factor rather than omitting it.
How does a transparent rubric map to a prioritized fix plan?
| Pillar | What we score | Evidence we verify | Weight |
|---|---|---|---|
| Get found (discoverability) | Catalog completeness (variants, GTIN, attributes); real-time price and stock; Schema.org and JSON-LD coverage; agent-readable policies and llms.txt | A live agent query returns the correct, current product set | 40 percent |
| Get bought (agent-buyable checkout) | Programmatic checkout with no human step; scoped payment credential support; MCP read layer and ACP checkout exposure; machine-readable order confirmation | A test agent completes a real, bounded purchase end to end | 35 percent |
| Stay in control (governance) | Spend cap per agent, transaction and period; approval gate with a step-up threshold; immutable order audit trail; signed mandate, authorization and execution record for finance | Finance can trace and defend a sample agent order start to finish | 25 percent |
Every criterion that scores below 3 becomes a line in a prioritized fix plan, ranked by weight and effort, and tied to the work that closes it. You get the score, the reason, and the next action, not a black box.
Is my store AI-ready?
Start with the numbers before any engagement. The free readiness scan runs the discoverability and checkout checks against your live store and returns a score across three pillars (can an agent see the store, can it act on it, can you host and control your own assistant) and the number of steps between you and an AI shopping assistant on your own store, so "is my store AI-ready" stops being a guess and becomes a list you can act on.
How do you make an existing store buyable by AI?
You do not rebuild the store. Saturnia Labs makes an existing shop buyable by AI agents with three layers that sit over what you already run on Shopify, Shopware or a custom platform.
- An agent-ready product feed: a clean, structured, machine-readable catalogue an agent can query, with no rebuild of your storefront.
- The agent interface: a thin MCP server over your store API, aligned with the ACP checkout standard, so an agent can search, price and buy.
- Guardrails: a spend cap, an approval gate above it, and a signed mandate logged to an audit trail your finance team can reconcile.
The stack is built on Anthropic's open commerce-agents blueprint, and we run the same three layers on our own storefront before we put them on yours.
How long does it take to make a store agent-buyable?
It depends on three things: catalog size and data quality, your platform and how it exposes a checkout, and how much of the governance layer already exists. A store with clean structured data and a modern platform moves faster than one rebuilding its feed. We size this before any build, so you commit to scope, not to a number we cannot yet stand behind.
Where do you start?
If an agent cannot find your products, cannot complete a purchase, or completes one your finance team cannot sign off on, you are not ready. Run the free readiness scan for the scored diagnosis, then close the gaps across all three pillars, governance included. Saturnia Labs is a member of the Claude Partner Network, the trust signal behind the governed, protocol-agnostic build we operate on our own storefront before yours.