Thirty years ago, buying online meant finding a store with a slow connection, trusting a remote seller, typing payment details, and waiting to see whether the order would actually arrive.
Today, a shopper can ask an AI assistant to compare products, check price and availability, remember preferences, and in some cases help complete the purchase.
The technology looks completely different. The pattern underneath it is surprisingly consistent.
Each major wave of e-commerce removed one difficult step from shopping.
And each time a step became easier, a new company or interface gained influence over discovery, payment, customer data, or the final purchase.
E-commerce history is easier to understand as a series of frictions that disappeared—not as a list of websites that became famous.
After reading this article, you should be able to explain the six major technology waves behind modern e-commerce, understand what AI shopping agents are actually changing, and use the same framework to ask what a creator, publisher, or merchant should prepare for next.
Wave 1: The Web Made the Store Reachable
Tim Berners-Lee invented the World Wide Web at CERN in 1989. By 1991, Web software was being released beyond the original laboratory environment, and in 1993 CERN made the Web software available on a royalty-free basis.[1]
The early Web was not built as a shopping mall.
It solved a simpler problem: a page could be published at an address and reached from another computer.
That was enough to change retail.
Amazon opened its online bookstore in 1995 and emphasized something a physical store could not easily match: a catalog of more than one million titles available through the Web.[2]
eBay also began in 1995 as AuctionWeb, connecting individual buyers and sellers who might never have found each other offline.[3]
The important change was not “books moved online.”
The shopper no longer had to travel to the store simply to see what was available.
The new control point was access to the digital catalog.
Wave 2: Search and Digital Payments Made Remote Buying Practical
Reaching a store was not enough.
The buyer still had to find the right seller and trust a remote payment.
PayPal was founded in 1998 and became part of the infrastructure that made digital payments more practical for consumers and online businesses.[4]
Search solved the discovery problem in a different way.
Instead of remembering every store address, a shopper could describe what they wanted.
Search advertising then connected that intent with merchants willing to pay for visibility near the query.
At the same time, e-commerce expanded through different regional models. Alibaba Group was founded in 1999, and Taobao launched in 2003 as a consumer marketplace in China.[5]
The second friction was reduced:
The shopper no longer needed to know the seller in advance or exchange payment in person.
Power moved toward search engines, marketplaces, digital identities, and payment systems.
Wave 3: Commerce Software Became Easier to Rent Instead of Build
Early online retail was technically demanding.
A merchant needed servers, a product database, a shopping cart, security, payment connections, order management, and people who could keep the system running.
Cloud infrastructure, software as a service, and APIs changed that.
Instead of building every component internally, merchants could combine specialized services.
Shopify’s origin illustrates the shift. Its founders began an online snowboard store in 2004 because existing e-commerce tools did not give them what they wanted. The Shopify platform itself launched publicly in 2006.[6]
The third friction was reduced:
A merchant no longer had to build every part of the commerce software stack alone.
That shifted influence toward cloud providers, commerce platforms, payment gateways, and software ecosystems.
Wave 4: Smartphones and Mobile Money Made Shopping Continuous
Desktop e-commerce still required the customer to sit at a computer.
The iPhone arrived in 2007, and Apple opened the App Store in 2008. Shopping, messaging, maps, payments, and media could increasingly happen on the same portable device.[7]
But mobile commerce did not develop in one global pattern.
In Kenya, M-Pesa launched in March 2007 and let people send and receive money through mobile phones, including many customers without traditional banking access.[8]
In China, marketplaces, mobile wallets, super-app behavior, short video, and livestreaming became deeply connected.
In Latin America, companies such as Mercado Libre combined commerce with payments, logistics, advertising, and merchant services.
The fourth friction was reduced:
Shopping and payment no longer depended on a desktop computer—or in some markets, even on a nearby bank branch.
The new control points included mobile identity, app distribution, payment, logistics, and local ecosystem data.
Wave 5: Content Became Part of the Store
Traditional search usually begins with a known need.
Recommendation feeds can create the need.
A shopper may discover a product in a short video, watch a creator test it, open the product page, and buy without starting a separate search.
TikTok describes this model as discovery commerce. In supported markets, TikTok Shop connects shoppable videos, livestreams, creators, product pages, checkout, and seller operations.[9]
YouTube, Meta, Naver, Pinterest, Amazon, and other platforms have built different versions of the same idea.
The fifth friction was reduced:
The shopper no longer had to stop the content, open a search engine, and rebuild the buying decision from the beginning.
Creators gained a new role because trust could now be connected directly with attribution and a transaction.
Figure 1. The regional paths were different, but each wave removed friction and created a new control point.
Wave 6: AI Is Starting to Do Part of the Shopping Work
Until recently, even a well-designed online store still expected the human to do most of the workflow.
The shopper searched, opened tabs, read reviews, compared specifications, checked availability, selected a product, entered payment details, and completed checkout.
Agentic commerce describes systems in which AI can perform some of those steps on the shopper’s behalf.
In 2026, this moved from demonstration to commercial infrastructure.
Google and Shopify: One Shared Language for Agentic Commerce
Google and Shopify co-developed the Universal Commerce Protocol, or UCP, an open standard for connecting AI agents with merchant systems across discovery, cart, checkout, orders, and post-purchase workflows.[10]
Google then introduced Universal Cart, designed to let shoppers collect products while using Google surfaces and complete checkout with participating merchants through Google Pay or retailer handoff.[11]
Shopify has also opened its Catalog and UCP tooling to developers. The practical idea is simple: product data can be structured once and made available to several AI shopping surfaces instead of building a custom integration for every assistant.[12]
ChatGPT: Shopping Inside the Conversation
OpenAI has developed a separate Agentic Commerce Protocol, or ACP, with Stripe and merchant partners.
ChatGPT shopping can now help users discover and compare products, and participating merchants can support buying flows directly from the conversation or through connected checkout experiences.[13]
UCP and ACP are different protocols.
That is worth remembering because the market is still developing. There is not yet one single technical route that every agent, merchant, and retailer uses.
Amazon: The Agent Can Shop Inside—and Sometimes Outside—Amazon
Amazon has been expanding Alexa for Shopping and agentic features such as Buy for Me.
Buy for Me can help purchase selected products from external brand websites when the item is not sold directly in Amazon’s store.[14]
This is another model: a retailer with an existing customer account, payment system, reviews, logistics network, and purchase history adds an AI interface on top.
What Does an AI Shopping Agent Actually Change?
The most important change is not that a chatbot can recommend a product.
Search engines and recommendation systems have done that for years.
The larger change is that one interface can begin to combine several jobs:
- understand a natural-language goal;
- find products across connected catalogs;
- compare attributes and prices;
- check live inventory and merchant terms;
- remember user preferences;
- prepare or complete checkout; and
- help with order or post-purchase questions.
A typical flow may look like this:
Human states the goal → AI narrows the options → Human reviews or approves → Merchant completes the transaction
The exact level of automation differs by platform.
Human approval, identity, payment permission, returns, and merchant responsibility still matter.
The New Friction: Can the Agent Understand the Product Correctly?
AI shopping creates a new problem for merchants.
It is not enough for a product page to look attractive to a human.
The information also has to be current and understandable to software.
That includes:
- product name and category;
- price and availability;
- size, material, color, compatibility, or other structured attributes;
- shipping and return rules;
- who the product is for;
- what problem it solves; and
- how it differs from similar products.
Shopify says its Catalog API turns product information into structured, queryable data for AI agents and keeps inventory and pricing synchronized across connected experiences.[12]
This question is already visible in merchant communities. Store owners are asking why products that rank well in traditional search do not appear in AI recommendations, how much catalog cleanup is needed, and how to measure sales attributed to agentic channels.
Those discussions are useful question signals. They are not proof that one optimization tactic will work for every store.
For Creators and Publishers, AI Shopping Changes the Decision Interface
A creator or publisher may not sell the product directly.
But they may influence the evidence that an AI agent uses when it decides what to recommend.
This makes several assets more important:
- original tests;
- clear comparison criteria;
- accurate specifications;
- fresh price or availability information when relevant;
- transparent commercial relationships;
- first-hand experience; and
- a reputation that exists beyond one platform.
If an AI agent can summarize ten generic “best product” articles, a rewritten specification list becomes easier to replace.
A test with visible evidence is different because the reader—or the system—can inspect where the conclusion came from.
Human Trust Does Not Disappear When AI Does More of the Shopping
AI can reduce the work of finding and comparing products.
It does not automatically remove uncertainty.
People still ask:
- Was this recommendation based on a real test?
- Is the seller reliable?
- Will the product fit my unusual situation?
- Who handles a return?
- Is the recommendation sponsored?
- Can I verify what the AI is telling me?
Recent consumer and merchant discussions show that shoppers often want human validation before committing money, especially when the decision is expensive or subjective.
For creators and publishers, that suggests a useful role: not competing with AI on summarizing basic facts, but providing evidence that helps the user verify a recommendation.
Why the Earlier Waves Still Matter
AI shopping agents do not replace the earlier e-commerce stack.
They depend on it.
In 2025, the International Telecommunication Union estimated that about 6 billion people, or 74% of the world’s population, were online.[15]
Agentic commerce also depends on:
- internet and mobile networks;
- cloud computing;
- merchant software;
- structured product catalogs;
- digital identity;
- payments;
- fraud prevention;
- inventory systems;
- shipping and returns; and
- customer permission.
AI sits on top of this infrastructure. It does not make it disappear.
A Simple Lens for the Next Commerce Technology
History gives us five questions that work better than trying to predict one winning company.
1. What Friction Gets Smaller?
Does the new system reduce search time, comparison work, uncertainty, payment steps, delivery time, or merchant setup cost?
2. Which Step Moves Out of Sight?
Does the user still need a search box, product page, cart, checkout screen, or separate retailer visit?
3. Who Gets the New Data?
Does the seller, creator, marketplace, AI assistant, payment company, or logistics provider learn more about the customer?
4. Where Does the Customer Start the Decision?
Search engine? Creator? Marketplace? Specialist website? AI assistant?
The starting interface often gains influence over what gets compared and what gets ignored.
5. What Becomes More Valuable Because of the Automation?
It may be structured product data, trustworthy reviews, original testing, payment permission, fulfillment reliability, or a direct customer relationship.
Figure 2. Use the history as a practical lens: identify the friction, the new data, the default interface, and the relationship that remains.
Run an Eight-Question Commerce Readiness Check
The original temptation is to turn this into a score.
A simpler method is more useful: answer Yes / Partly / No, then choose the two weakest areas to improve first.
- Discovery: Can people find you through more than one platform?
- Trust: Do you publish evidence that is harder to replace with a generic summary?
- Structured information: Can people and machines understand your product or knowledge accurately?
- Attribution: Can you connect content with a useful next action?
- Transaction path: Is there a clear way to move from decision to purchase, signup, or another useful action?
- Reachable audience: Can you reach the reader or customer again without winning the feed tomorrow?
- Revenue diversity: Can one weak revenue source fall without damaging the whole business?
- AI readiness: Are your important facts, rules, products, and workflows clear enough to be used by software without guessing?
The total number of “Yes” answers is not the point.
The useful result is knowing what to fix next.
If trust is weak, publish one original test. If structured data is weak, clean the product information. If the audience relationship is weak, give readers a reason to return directly. If attribution is weak, measure the next useful action rather than pageviews alone.
What the Next 30 Years May Change
AI shopping agents are not the final wave.
Future commerce may combine AI with augmented-reality fitting, autonomous delivery, persistent personal agents, machine-to-machine purchasing, local manufacturing, and new identity or payment systems.
We do not need to know exactly which interface will dominate.
The history gives us a more durable method.
Find the friction that disappears. Then ask who receives the data and who still owns the customer relationship.
The Main Idea
The Web made stores reachable.
Search and digital payments made products easier to find and buy remotely.
Cloud software made stores easier to build.
Smartphones made shopping continuous.
Creator platforms placed shopping inside content.
AI agents are now trying to remove part of the shopping workflow itself.
None of these waves erased the previous one. Each one depended on what had already been built.
For creators, publishers, and merchants, the useful lesson is not to chase every new feature.
It is to notice which friction is disappearing, which interface is becoming the starting point, and which asset you still control after the interface changes.
Continue Reading
- From Attention to Transactions: Why the Internet’s Business Model Is Changing
- The Race to Make Every Video Shoppable: YouTube, TikTok, Meta, and Naver
- Affiliate Marketing vs. Advertising: Which Model Wins the Creator Economy?
- Is Google Still an Advertising Company? Search, YouTube, AI, and Commerce
- Who Wins When Content Becomes Commerce? Platforms, Creators, and Publishers
Key Terms
- friction: a cost, delay, difficulty, or uncertainty that makes shopping or selling harder
- control point: a stage where control over access, data, payment, or infrastructure creates economic leverage
- agentic commerce: commerce in which AI software performs some shopping tasks for the user
- UCP: Universal Commerce Protocol, an open standard co-developed by Google and Shopify for agent-to-merchant commerce workflows
- ACP: Agentic Commerce Protocol, an open commerce protocol developed by OpenAI with Stripe and merchant partners
- structured product data: product information organized so software can reliably understand price, availability, attributes, policies, and variants
- merchant of record: the business legally responsible for the sale to the customer
- default decision interface: the place where the shopper normally begins comparison and selection
- direct audience: readers or customers you can reach through channels such as email, direct visits, membership, or customer accounts rather than only through a recommendation feed
Sources
- CERN — A short history of the Web
- Amazon — World’s Largest Bookseller Opens on the Web, October 1995
- eBay — Our History
- PayPal — Corporate history
- Alibaba Group — Introduction and milestones
- Shopify — About us and company milestones
- Apple — The App Store turns 10
- Vodafone — M-Pesa marks 15 years
- TikTok — TikTok Shop expands across Europe, May 2026
- Shopify / Google — Universal Commerce Protocol
- Google — Universal Cart, May 19, 2026
- Shopify — Agentic commerce for every developer, June 17, 2026
- OpenAI — Powering Product Discovery in ChatGPT, March 24, 2026
- Amazon — Generative and agentic AI in shopping
- ITU — 2025 Internet-use statistics
Status checked October 1, 2026. Agentic-commerce protocols, checkout availability, market coverage, attribution, and product eligibility are still changing. Community examples were used to identify practical questions, not as market benchmarks.