Agentic Commerce: When AI Starts Shopping for You

What Is Agentic Commerce?

Agentic commerce is a model of digital commerce where AI agents assist with, or act on behalf of, consumers throughout the shopping journey.

An agent may help:

discover products
compare options
evaluate specifications
check pricing
consider personal preferences
prepare a cart
start checkout
assist with completing a purchase.

Traditional ecommerce assumes that the customer manually performs most of those steps.

Agentic commerce allows parts of the process to be delegated.

Imagine someone looking for a laptop.

Instead of searching through dozens of product pages, the customer could say:

Find me a lightweight laptop for office work and software development. I need at least 16 GB of RAM, strong battery life, and I do not want a gaming laptop.

The agent interprets several constraints at once.

It can then eliminate unsuitable products, compare the remaining options, and present a shortlist.

With the right permissions and integrations, the journey may continue into cart and checkout.

The important shift is that the customer no longer needs to translate their entire need into a sequence of search queries.

They can express an intent.

Agentic Commerce Is More Than an Ecommerce Chatbot

Ecommerce chatbots are not new.

They have long been used to answer common questions such as:

Where is my order?

What is your return policy?

Does this product come in another colour?

Agentic commerce has a broader scope.

The system is not only responding to questions.

It can work through several steps toward a shopping objective.

For example:

I need a birthday gift for someone who recently started making coffee at home. My budget is $75.

An AI shopping agent could:

understand the occasion
→ explore several product categories
→ compare grinders, brewers, and accessories
→ stay within budget
→ check availability
→ recommend a combination
→ prepare the cart.

The customer provides intent.

The agent handles more of the discovery process.

From Search Queries to Shopping Intent

Traditional search requires customers to translate what they want into keywords.

Someone looking for comfortable walking shoes might search:

best walking shoes

then:

walking shoes for heel pain

then:

New Balance vs Hoka walking shoes

Each search refines the decision.

An AI shopping agent can receive the entire context at once.

For example:

I walk about five kilometres a day, occasionally have heel pain, prefer lightweight shoes, and want to spend less than $150.

This changes the competitive environment.

Ecommerce has traditionally competed around:

Which page ranks for the keyword?

Agentic commerce introduces another question:

Which product best matches the customer's intent according to the agent?

That is a fundamental change in product discovery.

AI Is Already Moving Into the Shopping Journey

Agentic commerce is not purely theoretical.

AI platforms are already moving beyond product recommendations toward commerce infrastructure.

OpenAI has been developing product discovery experiences inside ChatGPT as well as the Agentic Commerce Protocol, or ACP, designed to connect AI shopping experiences with merchant systems.

Google has also introduced the Universal Commerce Protocol, or UCP, as a common language intended to connect commerce platforms, AI agents, and merchants across activities such as discovery and checkout.

The direction is increasingly clear.

AI commerce is moving from:

help me choose

toward:

help me buy.

Can AI Already Buy Things on Its Own?

Not in every situation.

Human approval is still important in many commerce experiences, especially around payment and higher-risk decisions.

Agentic commerce is better understood as a spectrum of autonomy.

At one end, AI only helps discover products.

Then it may create a shortlist.

At a higher level, it can prepare a cart or checkout.

Eventually, certain low-risk and repetitive purchases may become increasingly autonomous.

Consider tasks such as:

reordering office supplies
replenishing household products
renewing recurring purchases
placing routine procurement orders.

These are easier to delegate than highly emotional or expensive purchases.

A customer may be comfortable allowing an AI agent to reorder detergent.

They may be much less comfortable allowing it to choose and buy a luxury watch without confirmation.

How Does an AI Shopping Agent Work?

A useful AI shopping agent needs several capabilities.

Understanding User Intent

The first task is understanding what the customer actually wants.

Not only the product category, but also:

budget
preferences
constraints
priorities
context of use.

For example:

Find me a hotel near my meeting location, under $120 per night, with breakfast, and suitable for a late-night check-in.

There are multiple requirements in one request.

The system needs to reason across all of them.

Discovering Products

Once the intent is clear, the agent needs access to relevant products or services.

That information may come from:

merchant feeds
marketplaces
APIs
product catalogues
websites
commerce platforms.

At this stage, merchant data quality becomes critical.

If pricing, variants, availability, or specifications are incomplete, the agent has less reliable information to work with.

Comparing Options

The agent may compare:

price
features
availability
shipping
reviews
return policies
compatibility
promotions.

The goal is not necessarily to show hundreds of choices.

One of the main benefits of AI shopping is reducing the number of decisions the customer needs to make.

Taking Action

Once the agent has identified a suitable option, it may:

add an item to the cart
request customer approval
choose an alternative when stock is unavailable
change quantities
prepare checkout.

As autonomy increases, permission design becomes increasingly important.

How Is Agentic Commerce Different From Traditional Ecommerce?

Most ecommerce experiences are designed around one assumption:

the primary user is a human.

The homepage is designed for humans.

Navigation is designed for humans.

Product pages are written for humans.

Search and filters are built for humans.

Agentic commerce introduces another type of user:

the AI agent.

An agent does not necessarily care whether the homepage has impressive animations.

It needs reliable information.

Price.

Stock.

SKU.

Variants.

Shipping details.

Compatibility.

Product identifiers.

Return conditions.

This creates a new design challenge.

Future ecommerce systems need to communicate effectively with people and machines.

Your Product Needs to Be Machine-Understandable

Traditional product pages can rely heavily on visual storytelling.

A brand might say:

Designed to move with you.

That might work as marketing copy.

It tells an AI agent almost nothing about whether the product suits a customer's needs.

The agent may also need:

material
dimensions
capacity
weight
compatibility
warranty
available variants
price
inventory
delivery options.

This does not mean brands should remove emotional storytelling.

Both layers matter.

People respond to stories.

Agents depend on data.

The strongest commerce experiences will support both.

Agentic Commerce Changes Product Discovery

Retailers have traditionally approached discovery through:

SEO
marketplace search
paid advertising
social media
recommendation engines.

AI shopping adds another interface.

Products can be discovered inside a conversation that did not initially look like a shopping query.

For example:

I am travelling to Japan in December. What kind of jacket should I bring?

The customer is discussing travel.

But the conversation creates a shopping need.

An AI agent may identify that need before the user ever opens a retailer website.

This moves product discovery further upstream in the customer journey.

Merchants may need to influence consideration before a traditional shopping session even begins.

GEO and Ecommerce Are Moving Closer Together

SEO has long helped retailers become visible through search engines.

As product discovery moves into generative interfaces, Generative Engine Optimization, or GEO, becomes more relevant to commerce.

A merchant no longer wants only to appear when a customer searches:

best headphones for travel

The merchant may also want product information to be considered when someone asks an AI assistant:

Which headphones are best for an eight-hour flight?

That means product information should be:

accurate
current
structured
consistent
verifiable.

In this environment, GEO and ecommerce increasingly overlap.

The AI needs enough information to understand not only that a product exists, but when it is a good fit.

Product Feeds Become More Strategic

Product feeds were once associated primarily with marketplace listings and shopping ads.

Agentic commerce gives them a broader role.

They may become one of the main ways merchants communicate product information to AI-powered commerce platforms.

Feed quality matters.

Product titles.

Prices.

Variants.

Inventory.

Images.

Identifiers.

Shipping details.

Availability.

Return information.

If these fields are inaccurate, the agent may recommend another product simply because it has better data.

As AI-driven shopping grows, catalogue management becomes part of AI visibility strategy.

Ecommerce Is Moving From Pages Toward APIs

Traditional ecommerce is heavily page-based.

An AI agent works more naturally with structured interfaces.

It may need to ask:

Is this item available?

Which variants are in stock?

What is the current price?

Can it be delivered to this address?

What does shipping cost?

What is the return policy?

Can checkout be initiated?

Commerce APIs make this type of interaction much easier than asking the agent to infer everything from a rendered webpage.

Protocols such as ACP and UCP point toward an ecommerce environment where machine-to-machine communication becomes a standard part of the shopping journey.

ACP and UCP Show Where Commerce Is Heading

As agentic commerce grows, common standards are becoming increasingly important.

Without them, every merchant would need a separate integration with every AI platform.

OpenAI's Agentic Commerce Protocol aims to support commerce interactions between AI agents, shoppers, and merchants.

Google's Universal Commerce Protocol serves a similar role across Google's commerce ecosystem.

These protocols can support areas such as:

product discovery
cart
checkout
payments
fulfilment
order status.

The standards are still evolving.

But they reveal an important direction.

Ecommerce is starting to prepare for machine customers.

Payments Are a Much Harder Problem

Finding a suitable product is relatively low risk.

Allowing an AI agent to spend money is more sensitive.

Once an agent can execute a transaction, several questions appear.

How much can it spend?

Which merchants can it use?

What happens if it purchases the wrong item?

How is user approval verified?

How are payment credentials protected?

How is fraud detected?

Agentic payments require clear authorization.

Giving an AI unrestricted access to a payment method would be a poor design.

Permissions can instead be constrained.

For example:

You may purchase approved office supplies from existing vendors up to $500 per month.

This is a much safer delegation model.

Trust Becomes a Machine-Readable Requirement

In traditional ecommerce, the customer chooses the merchant.

In agentic commerce, AI increasingly participates in that choice.

That creates two layers of trust.

The customer needs to trust the AI.

The AI needs reliable evidence that it can trust the merchant.

A merchant with:

inconsistent product data
unclear return policies
outdated stock
frequent pricing discrepancies

will be harder for an agent to use confidently.

Trust therefore becomes increasingly machine-readable.

Clear policies, verified catalogue data, fulfilment reliability, and accurate transaction information may all matter more.

Brand Still Matters

It is easy to assume that AI shopping will reduce every product to a comparison table.

That risk is real in highly commoditized categories.

If customers care only about:

price
specification
delivery time

AI may make comparison easier and reduce brand differentiation.

But consumers still have preferences.

Someone may ask:

Find me a Lenovo laptop suitable for software development.

The brand has already shaped the intent.

Brand building therefore remains important.

The difference is that a brand's value proposition needs to be understood not only by the customer but also by the agent.

Will Marketplaces Disappear?

Probably not.

Marketplaces already possess several assets that are extremely valuable to AI commerce:

large catalogues
merchant networks
reviews
payments
inventory data
fulfilment infrastructure.

Agentic commerce may change how users reach those systems rather than eliminating them.

Instead of opening a marketplace first, customers may begin with an AI assistant.

The marketplace becomes infrastructure behind the interaction.

Competition may gradually shift from:

Who owns the largest storefront?

toward:

Who has commerce infrastructure that agents can use most effectively?

Agentic Commerce Could Be Even More Important for B2B

Agentic commerce is often discussed through consumer shopping.

B2B commerce may eventually be an even stronger fit.

Procurement contains many repetitive activities.

Finding suppliers.

Comparing pricing.

Checking inventory.

Reviewing specifications.

Creating purchase requests.

Obtaining approval.

Issuing purchase orders.

An AI procurement agent could handle many of these steps.

Imagine the instruction:

Find an approved supplier for this cable specification. Prioritize existing vendors. If the price differs by more than five percent from our contracted rate, request approval.

The agent can operate within a business policy.

This kind of controlled purchasing is particularly compatible with automation.

The Procurement Agent May Become a New Type of Buyer

Most digital commerce today is optimized for human buyers.

Businesses may increasingly need to consider another audience:

buyer agents.

A procurement agent might know:

approved suppliers
budget limits
minimum inventory levels
contract pricing
payment terms
delivery SLAs
company policies.

When inventory drops, the agent could begin sourcing automatically.

Humans would only become involved when approval or negotiation is needed.

For B2B suppliers, this changes the value of digital presence.

A professional website remains useful.

But machine-readable catalogues, pricing rules, specifications, and procurement information become increasingly strategic.

What Should Ecommerce Businesses Do Now?

Not every retailer needs to build an AI shopping agent today.

But several foundations are already worth improving.

Improve Product Data

Make sure product information is complete and consistent.

Avoid:

different prices across channels
outdated inventory
unclear variants
duplicate SKUs
missing specifications.

If a human cannot confidently understand the catalogue, an AI agent will struggle too.

Combine Product Facts With Brand Storytelling

Marketing copy is still important.

But product pages also need factual information.

Include:

materials
dimensions
compatibility
capacity
warranty
delivery
returns
technical specifications.

The agent needs enough information to determine whether the product matches the user's intent.

Treat APIs as Strategic Infrastructure

Custom ecommerce platforms should begin treating commerce APIs as strategic assets.

Clear interfaces for:

catalogue
inventory
pricing
cart
checkout
orders
customers

make it easier to connect the platform with future AI commerce systems.

Invest in GEO

Product discovery is increasingly happening through generative AI.

Buying guides, product comparisons, category content, product specifications, reviews, and structured catalogue information can all help AI understand the merchant's products.

SEO remains important.

GEO becomes another layer.

Design Permissions Before Autonomy

If AI agents eventually gain transactional access, permissions should be designed carefully.

Use:

scopes
approval rules
transaction limits
audit logs
role-based access.

Autonomy needs boundaries.

Do Not Add AI Simply Because It Is Trending

Agentic commerce has significant potential.

That does not mean every ecommerce company needs its own shopping agent.

Start with customer friction.

Where are customers struggling?

Is the catalogue too large?

Are customers unsure which product fits their needs?

Are specifications difficult to compare?

Are repeat purchases common?

Does procurement involve excessive manual work?

The use case should come first.

AI should solve a real problem rather than become an additional feature with no clear value.

Will Agentic Commerce Replace Ecommerce?

No.

A better way to understand agentic commerce is as a new interface on top of existing commerce infrastructure.

The catalogue still exists.

Inventory systems still matter.

Order management still matters.

Payments still matter.

Logistics still matter.

What changes is how the customer interacts with those systems.

Today, the flow often looks like:

customer → website → ecommerce platform

A growing number of journeys may eventually look like:

customer → AI agent → commerce platform

Merchants with strong underlying infrastructure will be better prepared for both.

Will People Really Let AI Shop for Them?

Not for every purchase.

Autonomy will depend on:

transaction value
category
risk
personal preference.

Routine purchases are easier to delegate.

For example:

household supplies
office products
subscriptions
repeat orders.

High-value or emotionally significant purchases will likely retain more human involvement.

Agentic commerce will therefore probably develop across several levels of autonomy rather than moving directly from manual shopping to completely autonomous purchasing.

Human approval will remain important.

Agentic Commerce Changes the Meaning of Conversion

Traditional ecommerce analytics often begin when the customer lands on a website.

Agentic commerce may move part of that journey elsewhere.

A purchase could originate from:

ChatGPT
Gemini
AI Mode
an enterprise procurement agent
another AI assistant.

The merchant website may no longer be the first interaction.

Businesses will need better visibility into:

which AI channel initiated the transaction
which product was recommended
what user intent triggered discovery
how well AI-originated customers convert
which AI channels create new customers.

Commerce analytics needs to evolve with the interface.

The Long-Term Future Could Be Agent-to-Agent Commerce

The next stage goes beyond a buyer interacting with a merchant site.

A buyer agent could interact directly with a seller agent.

The buyer agent says:

I need one hundred units of this component delivered before Friday. Find the best available price.

The seller agent responds with:

inventory
volume pricing
delivery estimate
payment terms.

Some negotiation could eventually happen without employees manually handling each message.

This is especially relevant to B2B procurement.

At that point, ecommerce becomes less like a collection of websites and more like a network of systems able to transact with one another.