Agentic commerce is shopping powered by AI agents that autonomously search, compare, negotiate, and purchase on your behalf—fundamentally transforming how commerce works.
Unlike traditional shopping, where you browse, click, and compare manually, agentic commerce delegates these tasks to an AI that can reason through your requirements.
"Find a birthday gift for my mom under $75—her birthday is March 15th, she likes gardening, and she hates anything too techy."
The agent parses the recipient (mom), occasion (birthday), date (March 15th), budget ($75), interests (gardening), and avoidances (tech), then builds a constraint profile.
It queries product catalogs, filters by category (gardening, home), excludes electronics, applies the $75 budget cap, and ranks options by fit.
It weighs products against your constraints—price, reviews, availability, shipping speed—and selects the best match.
It completes the purchase using authorized payment methods, within spending limits you've set.
"We chose a ceramic planter set ($68) with high ratings from gardening enthusiasts. It's gift-wrapped and arrives two days before her birthday."
McKinsey identifies three interaction models emerging in agentic commerce:
Your personal agent talks directly to a merchant's website or API. It browses, compares, and purchases just like you would—but faster and with perfect recall of your preferences.
Your agent negotiates with a retailer's agent. They can haggle on price, arrange bundle discounts, or coordinate complex transactions that would take humans hours.
Intermediary platforms coordinate multiple agents across multiple merchants—orchestrating complex purchases like travel itineraries or home renovation projects.
This isn't just "better chatbots." Agentic commerce represents a fundamental shift in who—or what—does the shopping.
| Traditional E-commerce | Conversational Commerce | Agentic Commerce | |
|---|---|---|---|
| Who acts | Human browses & clicks | Human chats, AI assists | AI agent acts autonomously |
| Discovery | Search bars, filters, ads | Chat-based Q&A | Agent reasons through data |
| Decision | Human compares options | Human decides with AI help | Agent decides within parameters |
| Execution | Human checks out | Human approves purchase | Agent completes transaction |
| Speed | Minutes to hours | Minutes | Seconds to milliseconds |
| Complexity | Limited by human patience | Limited by conversation | Handles dozens of constraints |
Conversational commerce helps you shop—agentic commerce shops for you. The agent doesn't just answer questions—it takes action, makes decisions within your boundaries, and completes transactions. You set the constraints; the agent handles the rest.
These aren't hypotheticals—they're the kinds of complex, constraint-heavy shopping tasks that agents handle today.
"Recommend a retinol serum for acne scars that's safe for sensitive skin and good for someone just starting retinol."
What the agent does:
"We chose a low-strength retinol because you mentioned sensitive skin. Start 2-3x per week and alternate nights with other actives. This product has strong reviews from users with similar skin concerns."
Why regular search struggles: Typing "retinol for acne scars sensitive skin" into a search box returns thousands of results with no reasoning about why one product is safer than another for someone just starting out.
"Can you put together a pregnancy-safe, fragrance-free dark-spot routine under $60 that gets here by Friday?"
What the agent does:
"Because you're pregnant, we skipped ingredients that aren't recommended and chose gentler options for dark spots instead. Everything is fragrance-free, arrives by Friday, and the full routine comes to $54."
Why regular search struggles: A normal search makes you do the work—you'd have to check pregnancy safety, fragrance-free labels, shipping speed, and price for each product, then figure out what routine actually makes sense. An agent applies all those filters at once and gives you a coherent plan.
"Reorder my usual moisturizer, but if the price increased more than 10%, find me an alternative."
The agent checks the current price against your last purchase, finds comparable products if needed, and explains the tradeoff.
"Find the best deal on this sunscreen across Amazon, Sephora, and Ulta—apply any coupons."
The agent queries all three retailers, applies available promo codes, factors in shipping, and returns the best total price with reasoning.
"Order my usual cleanser—if it's out of stock, find the closest match."
The agent checks inventory and, if it's unavailable, finds products with matching attributes (same skin type fit, similar ingredients, comparable reviews) and explains why.
This isn't a future concept—major technology companies and focused innovators are deploying live products and protocols today.
Agentic Commerce Protocol (ACP) — Powers ChatGPT Instant Checkout. Users can complete purchases directly in chat without leaving the conversation.
Agent Payments Protocol (AP2) — Open protocol backed by Mastercard, PayPal, and Amex. Uses cryptographic mandates for verifiable, auditable transactions.
Agent Commerce Kit (ACK) — Open-source toolkit addressing why legacy payment rails fail for agents: they're too slow, too expensive for micropayments, and require human authorization. ACK provides wallets, verified identity (ACK-ID), and spending rules (ACK-Pay)—the financial primitives agents actually need.
Model Context Protocol (MCP) — Standard for AI to connect with external tools and data sources. The "USB for AI integrations."
Buy with Pro — In-chat shopping experience integrated with PayPal and Venmo. Search leads directly to purchase without leaving the interface.
AI-ready cards and Agent Pay — Payment networks building infrastructure for agent transactions, including tokenized credentials and programmable spending limits.
While tech giants build end-to-end platforms, open-source initiatives like Catena Labs' ACK provide the foundational building blocks—identity, wallets, transaction rules—that let any developer build agentic commerce capabilities.
For agents to transact across platforms, they need shared standards—much like HTTP enabled the web. Five key protocols are emerging.
Catena Labs — Open-source financial primitives for agents: verified identity (ACK-ID), wallets, and programmable spending rules (ACK-Pay). The base layer that lets agents hold funds and transact.
OpenAI + Stripe — Enables agents to discover merchant capabilities, browse products, build carts, and complete checkout. Powers ChatGPT's instant purchasing.
Google — Payment-agnostic protocol using cryptographically signed mandates. Creates auditable trails linking user intent to agent actions. Backed by major payment networks.
Anthropic — Universal standard for AI to access external resources, tools, and data. Enables agents to connect with any system that implements the protocol.
Open Standard — Allows agents from different platforms to discover each other, exchange capabilities, and coordinate actions. The foundation for agent negotiation and collaboration.
These protocols are still evolving, and the industry hasn't consolidated around a single standard. But the direction is clear: agents need interoperable ways to discover, transact, and verify—and the infrastructure is being built now.
When your next customer is an AI agent, your infrastructure needs to change. McKinsey identifies six key domains.
Agents need structured, semantic data they can reason about—not marketing copy. Ingredients, specifications, use cases, and compatibility signals must be machine-readable.
Stale data breaks agent trust. If an agent recommends something that's out of stock, users lose confidence. Real-time accuracy is table stakes.
Agents need programmatic access, not human-facing websites. Product discovery, cart management, and checkout must be API-accessible.
How do you verify an agent is authorized to buy? New frameworks like "Know Your Agent" (KYA) are emerging alongside traditional KYC.
Agents need to know delivery times, shipping costs, and return policies to make informed decisions. These must be exposed programmatically.
If your loyalty program isn't API-accessible, agents can't apply benefits. Personalization data needs to flow to agent systems.
Agentic commerce isn't without challenges. Honest acknowledgment of risks builds trust—and the industry is actively addressing them.
Who's responsible when an agent exceeds its intended scope? Protocols like AP2 address this with cryptographic mandates and spending limits, but frameworks are still maturing.
Agents can make mistakes or misinterpret intent. Robust fail-safes, human-in-the-loop review for high-value transactions, and clear rollback mechanisms are essential.
Some regions are more cautious about automated payments. In Germany and Japan, many consumers prefer traditional payment methods. Adoption will be uneven.
When an agent-driven purchase goes wrong, who's liable? The AI platform? The merchant? The user? Legal frameworks are still catching up to the technology.
These are solvable problems. The industry is addressing them through spending limits, audit trails, human approval for significant purchases, and evolving regulatory frameworks. The direction is toward more trust, not less.
"This is not a wait-and-see moment. Before long, nearly all retailers will have to grapple with the fact that a significant percentage of their customers will not be human users, but rather, AI agents."
— Lareina Yee, McKinsey Senior Partner
McKinsey notes this transformation can move faster than prior e-commerce revolutions because agents can "ride on the rails" of existing digital infrastructure. The 5.6 billion people already connected to the internet don't need new hardware—just new software.
Whether you're a merchant preparing for agent traffic or exploring how to build agentic experiences, the transformation is underway.
Want to dive deeper? Explore our Resource Hub for protocol documentation, analyst reports, and the latest developments.
Why vector search isn't enough—and what comes next. Our thesis on decision geometry and the NoVEC moment.
Agentic commerce is AI-powered shopping where intelligent agents autonomously search, compare, negotiate, and purchase products on your behalf. Unlike chatbots that just answer questions, agents take action—completing multi-step transactions within parameters you set.
Conversational commerce uses AI to assist humans who still make decisions and complete purchases. Agentic commerce goes further: the AI agent makes decisions and executes transactions autonomously within your boundaries.
Yes. With protocols like ACP and AP2, agents can complete real purchases using cryptographic verification and spending limits to ensure they only act within authorized boundaries.
Structured product data that agents can reason about—real-time inventory, clear pricing, ingredient/specification details, shipping options, and return policies. The more semantic and machine-readable, the better.
Emerging protocols use cryptographically signed mandates linking user intent to agent actions. Users set spending limits, approve merchant categories, and define constraints. Every transaction creates an auditable trail.
Authorization errors (agent exceeds scope), cascading failures, and accountability gaps. The industry is addressing these through spending limits, human-in-the-loop approvals, and detailed audit trails.
It's already beginning. ChatGPT Instant Checkout, Perplexity Buy with Pro, and Google's AP2 are live. McKinsey projects $3-5 trillion globally by 2030, with 44% of AI search users already preferring it.
Three priorities: (1) Structure product data for machine readability, (2) Ensure real-time accuracy for inventory and pricing, (3) Expose key information via APIs—agents need programmatic access.