Agentic Commerce for Product Managers: What to Build When AI Agents Do the Shopping
TL;DR
Agentic commerce is the shift from consumers browsing and buying to AI agents researching, comparing, and purchasing on their behalf. ChatGPT now handles 50 million shopping queries daily. Amazon's Rufus serves 300 million users. The protocols that govern how agents discover, compare, and transact are already being standardized by Stripe, OpenAI, Google, and Anthropic. Product managers building anything in retail, SaaS, or marketplaces need a strategy for the agent channel now. This guide covers what agentic commerce actually is, the infrastructure layer being built around it, and where AI PMs should focus their roadmaps.
The AI PM Minute
One tactic to make you a sharper AI PM, twice a week. 60 seconds to read. Free.
No fluff. Unsubscribe anytime.
What Agentic Commerce Is and How It Differs From AI Search
AI-assisted search is when a user asks a question and gets a recommendation. Agentic commerce is when the agent completes the purchase. The distinction matters for product strategy because the two channels require entirely different investments.
In AI search, your product surfaces well in ChatGPT or Perplexity's answer. In agentic commerce, your product needs to be discoverable by an agent, negotiable by an agent, and transactable by an agent without a human in the flow. The agent is your new customer interface. Gartner predicts that 40% of enterprise applications will embed AI agents by 2026, and a mid-2025 Gartner survey shows 62% of enterprises are already experimenting with purchasing agents.
Traditional AI-assisted search
- 1.User asks ChatGPT which laptop to buy
- 2.AI recommends options with reasoning
- 3.User clicks through to retailer website
- 4.User completes purchase manually
- 5.Role of AI: discovery and referral
Agentic commerce
- 1.User tells agent: 'Buy the best laptop under $1,500 for video editing'
- 2.Agent searches multiple stores, compares specs, checks reviews
- 3.Agent uses stored credentials to complete purchase
- 4.Agent notifies user of completed order
- 5.Role of AI: autonomous purchasing decision-maker
The current state: agents can discover and recommend, and in specific verticals (OpenAI ChatGPT Shopping, Google AI Mode in Search) they can complete transactions for supported merchants. The infrastructure to make this universal is being built now by the major platforms.
The Three Protocols That Will Define Agentic Commerce
Three protocol layers are being standardized in 2026 that will determine how agents interact with commerce products. AI PMs need to understand each because they require distinct engineering investments and have different adoption timelines.
Agentic Commerce Protocol (ACP) by Stripe and OpenAI
What it does: ACP enables agents to discover products, get pricing, and complete purchases without a human in the loop. It provides a structured API layer that merchants implement so any compatible agent can transact with their store. ChatGPT Shopping already runs on ACP for supported merchants.
PM implication: Implementing ACP gives your products access to ChatGPT Shopping's 50 million daily queries. Stripe handles the payment infrastructure. Engineering effort: 2 to 4 weeks for an experienced team. Early ACP implementers get merchant page indexing before competition intensifies.
Google's Universal Commerce Protocol (UCP)
What it does: UCP covers the full commerce journey from discovery through post-purchase, designed to work across Google's AI Mode in Search, Google Shopping, and third-party agents. It standardizes how product data, pricing, inventory, and checkout flows are exposed to AI systems.
PM implication: Google still controls significant purchase intent discovery. UCP compliance means your products appear in Google's AI-driven commerce flows. The spec was published in May 2026 and adoption guidance is live in Google Merchant Center. Watch the Q3 2026 tooling roadmap before committing engineering resources.
Anthropic's Model Context Protocol (MCP) for Commerce
What it does: MCP provides the real-time data connectivity layer: live inventory, pricing changes, shipping estimates, and account-level personalization. Where ACP and UCP define transaction completion, MCP powers the informational substrate that agents query before and during a purchase decision.
PM implication: An MCP server that exposes accurate, structured product data makes your products more likely to be selected in agent comparison flows. This is currently the most underinvested opportunity among mid-market companies. Build cost is relatively low; compatibility is broad across all major models.
What Changes in Your Product Stack
Agentic commerce does not just add a new channel. It changes what good product data means, how you think about pricing logic, and what your checkout flow looks like when there is no browser session involved.
Product data completeness
Agents make purchase decisions on structured data, not page design. Incomplete product attributes, missing specifications, or vague descriptions get downranked or skipped. The new baseline: every product needs complete attribute coverage with machine-readable values. 'Large' is not a size; 'XL (46cm chest)' is.
Pricing and inventory APIs
Agents check price competitiveness in real time across multiple merchants. Your pricing API needs to return accurate real-time prices, including discounts and availability, with sub-200ms response times. Stale inventory data causes failed agent transactions, which directly damages your trust scores with agent platforms.
Headless checkout
Agentic purchases do not go through your cart UI. The ACP transaction layer handles checkout server to server. Your checkout logic needs to be fully API-accessible independent of your web frontend. This is a significant engineering investment if your checkout is tightly coupled to your frontend.
Trust and safety for autonomous transactions
When an agent makes a purchase without user review of each step, your fraud detection, order verification, and dispute resolution processes need to handle non-human sessions. Agent purchases have different fingerprints than human browsing and will trigger existing fraud rules unless you explicitly model for them.
Attribution and analytics
Current analytics models assume a browser session. Agent purchases arrive via API with a different metadata structure. You need new attribution models for the agent channel to understand conversion, AOV, and LTV for agentic customers versus direct buyers.
Build Strategy for the Agent Channel
The AI PM Masterclass covers agentic product strategy, distribution channel shifts, and how to roadmap for the emerging agent ecosystem, taught live by a Salesforce Sr. Director PM.
Highest-Leverage Roadmap Investments
Not everything needs to be rebuilt for the agent channel at once. These investments are ranked by time-to-impact and defensibility.
Agent-readable product data
Enrich your product catalog with complete, structured, machine-readable attributes. This is the foundation everything else depends on and it directly improves AI search ranking today, before you implement ACP or UCP.
ACP integration via Stripe
Implementing ACP gives you access to ChatGPT Shopping's 50 million daily queries. Stripe handles the payment infrastructure. Engineering effort: 2 to 4 weeks for an experienced team. Early movers get indexed before competition intensifies.
MCP server for your catalog
A well-structured MCP server exposes live inventory, pricing, and product specifications to any MCP-compatible model. Anthropic, OpenAI, and Google all support MCP. Broad reach at relatively low build cost.
Agent-channel pricing experiments
Agents optimize on stated buyer criteria (price, specification fit, delivery speed) rather than anchoring to original prices or responding to visual urgency cues. Dynamic pricing models may perform differently in agentic versus human channels. Measure before assuming parity.
Build Now vs. Watch in 2026-2027
The agentic commerce stack is maturing fast. Some pieces are ready for production investment; others are still in flux.
Build now
Structured product data enrichment
Immediate payoff in AI search and AI Mode in Google. No protocol dependency. Improves ranking in every agent channel that already exists.
ACP integration via Stripe
ChatGPT Shopping is live and growing. Stripe handles the hard parts. Early implementers get merchant page indexing before competition intensifies.
MCP server for real-time catalog data
Low build cost and broad compatibility across all major models. Currently the most underinvested opportunity among mid-market companies.
Headless checkout API
If your checkout is frontend-coupled, this is the highest-risk gap to close. Agent transactions cannot complete without it.
Watch and prototype in H2 2026
Agent loyalty and retention models
How do you retain an agentic customer who has no brand loyalty and optimizes on price and spec? New models are needed, ones that reward the agent and through it the buyer for repeat behavior.
Agent trust scoring programs
Platforms are building trust scores for merchant agents and buyer agents. Early participation may carry a network advantage similar to early Google seller ratings.
Post-purchase agentic experiences
Returns, support, reorders: the agentic flow extends past checkout. The winning commerce products will own the full agentic lifecycle, not just the transaction.
UCP compliance
Google's UCP spec is live but adoption tooling is still maturing. Monitor the Q3 Merchant Center roadmap before committing engineering resources.
Navigate the Agent Channel Before Your Competitors Do
The AI PM Masterclass covers agentic strategy, distribution shifts, and the infrastructure decisions that determine who wins in the agent channel, taught live by a Salesforce Sr. Director PM.
Related Articles
Before you go: get the AI PM Minute
One tactic to make you a sharper AI PM, twice a week. 60 seconds to read. Free.
No fluff. Unsubscribe anytime.