I've Watched Five Waves Reshape Ecommerce. The Sixth Is Following the Same Script.

I spent more than 20 years building and operating an ecommerce business, starting when many retailers still treated the website as a side channel. We built from zero into the No. 310 retailer among North America’s largest online retailers by web sales (Digital Commerce 360, 2023) in one of the hardest categories to sell online.

Over that time, five waves fundamentally changed how customers found products, trusted information and completed purchases: marketplaces, mobile, personalization, the Prime-driven omnichannel fulfillment model and social commerce.

Each looked different on the surface, but the pattern underneath was remarkably consistent. Customer behavior started moving before most retailers believed the shift would matter. The infrastructure showed up before the volume did. Then a traffic, cost or convenience curve changed. The retailers that responded by removing friction usually gained ground. The ones that treated it as one more channel to manage usually ended up reacting later.

I think we are seeing that same pattern again with agentic commerce.

The five waves that built modern ecommerce

1. Marketplaces turned selection and trust into a platform

Early ecommerce sites were digital versions of individual stores. Marketplaces combined products, sellers, reviews, payment protection and fulfillment in one transaction environment. The breakthrough was not simply more assortment. It was enough trust to buy from a seller the customer had never heard of. Today, Amazon says independent sellers account for more than 60% of sales in its store. The long-term advantage came from owning the environment in which selection, confidence and checkout came together.

2. Mobile removed shopping from a specific place and time

The smartphone did not create a new desire to shop. It removed the need to wait until someone was sitting at a computer. It also exposed every unnecessary step retailers had built into the desktop experience. By 2025, mobile represented more than half of full-year online spend for the first time and 56.4% of holiday ecommerce revenue. The mistake was treating mobile as a smaller website instead of a different customer environment.

3. Personalization changed the catalog from fixed to individual

As assortments expanded, the customer's problem shifted from access to relevance. Recommendations, predictive search, behavioral signals and loyalty data began ranking the catalog differently for each shopper. The strongest retailers understood that the value was not the recommendation widget. It was the compounding loop in which every interaction improved the next one.

4. The Prime effect turned fulfillment into part of the product

Customers stopped separating the product from the promise around it. Availability, delivery date, pickup, installation, returns and service became part of the decision. Stores became inventory nodes, pickup points and return centers. The winners connected inventory, order management, labor and customer communication around one promise. They did not bolt curbside pickup onto the website and call it omnichannel.

5. Social commerce separated discovery from the storefront

Traditional ecommerce assumed the customer intentionally began a shopping trip. Social and creator commerce changed the sequence. The product began finding the customer. During the 2025 Prime Day period, Adobe measured affiliates and partners, including influencers, at 19.9% of online revenue, while influencer-referred shoppers converted at ten times the rate of social media overall. Discovery had moved somewhere the retailer did not control.

The four-stage script behind every wave

Stage 1: The behavior proves out somewhere "over there" and gets dismissed as niche. Japan's i-mode ecosystem showed consumers would use phones for commerce years before the iPhone. China's livestream shopping market proved entertainment and checkout could merge before the model reached the United States. Online auctions, review systems and comparison engines showed the marketplace model before most traditional retailers saw the full implication.

Stage 2: An infrastructure layer quietly ships. The App Store, Fulfillment by Amazon, modern order-management systems, stored payment credentials and social checkout all arrived before demand reached its final scale. I learned to pay attention when the infrastructure showed up before the volume. Platforms and payment networks do not build this much plumbing unless they expect it to be used.

Stage 3: A traffic or cost curve bends while incumbents debate whether the behavior is real. Mobile traffic kept rising while retailers argued about whether customers would complete purchases on a phone. Social discovery accelerated faster than traditional merchandising calendars.

Stage 4: The work spills out of digital and into the operating model. At that point, inventory, pricing, fulfillment, service, measurement and accountability all have to line up around the customer experience. Companies that leave the change inside a channel team usually end up chasing customer expectations instead of setting them.

If I were running a retail P&L today, I would pay less attention to agentic commerce's current share and more attention to three things: how much infrastructure is being built, how quickly customer behavior is moving and how much friction the technology removes.

Agentic commerce is moving into stage three

That is why the last year stands out to me.

The infrastructure layer is no longer theoretical; much of it is now in production. Visa introduced its Trusted Agent Protocol in October 2025 so merchants can distinguish legitimate agents from malicious bots. Google launched the Universal Commerce Protocol at NRF in January 2026 and introduced Universal Cart in May. Microsoft began rolling out Copilot Checkout in January with PayPal, Shopify and Stripe. Mastercard launched Agent Pay for Machines in June. By fall, this had moved past protocol announcements: Salesforce lists UCP-powered Google checkout for general availability in October, and Shopify says millions of merchants can sell through AI chats.

I do not see those as isolated experiments. When the card networks, Google, Microsoft and commerce platforms all build transaction rails in the same window, I pay attention. That is the kind of signal I would have wanted my team watching in earlier ecommerce shifts.

The customer side is moving too. Adobe measured AI-driven retail traffic up 138% year over year in May 2026 alone. AI-referred shoppers converted 54% better and generated 53% more revenue per visit than non-AI traffic. During the 2026 Prime Day event, AI shoppers converted 40% better. A year earlier, AI traffic was still converting materially worse than traditional channels. More recently, Shopify's 2026 holiday survey found 65% of shoppers expect to use AI for at least one shopping task this season. That is intent, not transaction share, but it tells me the behavior is moving beyond early adopters. More research and comparison is happening before the shopper reaches the retailer.

Salesforce reported that AI and agents influenced 20% of global online holiday sales in 2025, representing $262 billion, and that retailers operating their own shopper agents grew sales 59% faster than retailers without them. Those are Salesforce's own measurements, so do not treat them as a census of retail. Treat them as another signal that AI is moving closer to the commercial decision.

The forecasts are still far apart. EMARKETER expects AI platforms to account for only about 1.5% of U.S. retail ecommerce sales in 2026, or roughly $20.9 billion. McKinsey estimates agents could mediate $3 trillion to $5 trillion in global consumer commerce by 2030. I do not see that as a contradiction. Early in a shift, today's share can be small while the infrastructure being built points to a much larger change. The question I would put in front of a leadership team is not, "How many sales are agents driving today?" It is, "Are the systems that shape discovery, comparison and transaction being rebuilt around them?"

I believe retailers have an 18- to 24-month window to get positioned. That does not mean every order will become autonomous. It means the retailers that build the right foundations now will learn while the traffic is still manageable, rather than trying to rebuild under peak-volume pressure later.

What the next five years look like if the script holds

1. The agent becomes the first decision layer

More shopping journeys will start inside an assistant the retailer does not control. That compresses discovery, comparison and shortlisting into a conversation. For merchandising teams, product data becomes much more important because an agent cannot recommend what it cannot understand. Titles, attributes, compatibility, variants, policies, inventory and delivery promises have to answer the customer's actual question, not simply fill a template.

2. Checkout becomes an API handshake backed by trust

The checkout page will remain, but more transactions will be prepared or completed through structured connections between agents, merchants and payment providers. From my experience leading ecommerce, the conversation is the easy part. The real work is preserving the right variant, price, promotion, loyalty benefit, tax treatment, delivery commitment and return rule. An agent that cannot reach the business logic behind the storefront is still a chatbot.

3. The answer gets monetized

Monetization always follows attention. Banner ads followed portals. Search ads followed Google. Retail media followed marketplace traffic. Promoted placement inside AI discovery is the predictable next act. Shopify has already described future promoted placements in Catalog-powered experiences, while Google's Direct Offers pilot is designed to present merchant deals inside AI Mode. Retailers will have to decide how far paid placement can go before it starts to hurt relevance and trust.

4. Product visualization becomes personalized simulation

Many categories remain difficult online because the customer cannot confidently imagine the outcome. Generative imagery, computer vision, virtual try-on, room visualization and digital product models reduce that uncertainty. Google now lets shoppers virtually try on clothing from billions of listings using an uploaded image. What ties this back to agentic commerce is the same foundation: accurate, structured product information and governed digital assets.

5. Replenishment and fulfillment become delegated and margin-aware

The next fulfillment wave is not just about faster delivery. More of the decisioning will happen in systems that choose what to reorder, where inventory should sit and how an order should move based on cost, speed, labor, service risk and customer rules. Replenishment is a logical place for delegated purchasing because the customer has already made the product decision. For repeat purchases, loyalty may start to look less like a program and more like a default setting. The agent that holds the trusted customer relationship can influence share of wallet before the customer opens an app.

Bonus prediction: agents become negotiators

One area I would watch closely is negotiation. Customers rarely care about sticker price alone. They care about the all-in deal: product price, delivery cost and timing, tax, return terms, financing, service and whether the order can arrive when promised. An agent could eventually take a request like, “Get me this product delivered by Friday for no more than $1,200, with free returns,” and shop those terms across multiple merchants. The retailer side is where this gets interesting. Pricing and offer logic would need to respond in real time to margin, inventory position, fulfillment cost, customer value and business rules. At that point, pricing is no longer just something displayed on a product page. It becomes something the retailer’s systems can negotiate within defined guardrails. I would not expect that to happen evenly across every category, but high-consideration purchases are a logical place to watch first.

What Salesforce and Shopify are doing about it

For most retailers, do not start by asking whether we should build every protocol integration ourselves. Start with two questions: what is our commerce platform already enabling, and are we operationally ready to use it?

Salesforce is building connection depth

Salesforce is positioning Agentforce Commerce as the connection layer between AI shopping experiences and the enterprise systems that govern the order. Dreamforce made that strategy more concrete. Salesforce now lists store-specific inventory support in Agentic Commerce Search as generally available, UCP-powered Google checkout for general availability in October, and Shopper Agent product comparisons as available now, with page awareness and authenticated shopper context scheduled for October. It is also pushing agents into operations through Commerce Operations MCP and Agent Plugins and an AI Migration Toolkit for Storefront Next.

Having led an enterprise migration to Salesforce Commerce Cloud, I see the logic. In complex retail, generating a helpful answer is the easy part. The difficult part is making sure that answer matches the pricing engine, inventory position, order-management system, fulfillment network, service policy and customer record. That is where I would want an enterprise platform spending its energy.

Do not assume that the platform solves the operating model. It does not decide product truth, consent, guardrails, attribution or ownership for the retailer.

Shopify is building distribution breadth

Shopify is taking a network approach. Agentic Storefronts give merchants centrally managed access to AI channels including ChatGPT, Microsoft Copilot, Google's AI Mode and Gemini, with Shopify continuing to add surfaces through UCP. Shopify Catalog structures and distributes product data, while the Universal Commerce Protocol, co-developed with Google, gives agents and merchants a shared language from discovery through checkout. Shopify has also opened its Agentic plan to brands whose primary store is not on Shopify.

Shopify says AI searches powered by Shopify Catalog convert at twice the rate of searches built from scraped product data. That is Shopify's own measurement. I would still take the underlying point seriously: controlled, current, structured product data should outperform information an agent has to reconstruct from a webpage.

Shopify is also pushing AI deeper into the merchant workflow. Its new Canvas design surface lets Sidekick work directly on theme files to build and refine storefronts, while SimGym lets merchants test storefront changes against simulated AI shoppers. For merchants, more of the work moves into the platform instead of another custom integration project.

The strategies are converging. Salesforce is adding standardized distribution through UCP; Shopify is moving AI deeper into merchant operations. Neither can clean the catalog, resolve internal ownership or make sure operations can deliver what the agent promises.

What ecommerce leaders should do now

Do not stand up an "agentic commerce team" and park this work there. It touches merchandising, technology, marketing, operations, legal, finance and customer service, so it needs to sit inside the operating plan.

1. Audit the products that matter most

Ask an AI assistant to compare your top 50 products with the closest competitive alternatives. Look for missing attributes, conflicting specifications, weak policy language, stale inventory signals, poor variant clarity and content that sounds persuasive to a human but says very little to a machine. Start where the revenue and margin are concentrated.

2. Make the promise executable

Trace what an agent would need to confirm price, availability, delivery, installation, pickup, returns, loyalty and service. Look for every point where that promise breaks between the storefront and the systems that have to fulfill it. An agent giving a confident answer that the operation cannot deliver is worse than no answer.

3. Measure influence, not only the last click

AI-assisted journeys will not fit cleanly into existing channel reports. Leaders need visibility into AI citations, referrals, product visibility, assisted conversion, revenue per visit, margin, returns and service contacts. Do not build another vanity dashboard around this. Find out whether AI is sending better customers and whether the economics still work after fulfillment and returns.

4. Set guardrails before scaling action

Define what an agent can do without approval and what requires a human. Most of the pain will show up in exceptions, so substitutions, price changes, promotion stacking, high-value purchases, delivery failures and returns need explicit rules. Also define how we monitor failures and test changes. Write the rules, then watch whether the agents follow them.

5. Name an owner for the end-to-end outcome

Enabling the platform does not solve ownership. Someone still has to own product truth, channel participation, customer consent, risk, measurement and the commercial outcome across functions. Every major ecommerce wave exposed the cost of fragmented ownership. Agentic commerce can expose it faster because bad data and broken promises can move across external AI surfaces at machine speed.

What retail leaders should take away

Every wave I lived through looked like a digital project at first and turned into a leadership challenge because it changed how customers found products, trusted information and completed purchases. Agentic commerce is changing all three at once.

Agent-completed sales are still small. The infrastructure and the shopping behavior behind them aren't. Anyone who has operated a holiday season knows that the fixes showing up in October needed to start months earlier. The same is true here. The work is catalog, systems and ownership work, and none of it gets easier under peak pressure.

The question I would ask the team this week is simple: if an AI assistant compared our most important products with our three closest competitors today, would it have enough clean, consistent and verifiable information to recommend us, and could our operation deliver the promise it made?

If the answer is "we're not sure," that is where I would start.

After 20 years, I still come back to the same test: does the technology make it easier for the customer to do business with you? The only thing changing is that an AI agent may now be doing part of the shopping.

About the author

Jeff Douglas

I bring 20+ years of ecommerce leadership experience, including revenue and budgeting responsibilities, customer experience and digital marketing leadership, Salesforce Commerce Cloud modernization and high-consideration retail. I advise businesses on customer experience, growth priorities, technology and AI-driven discovery.

Experience