Agentic AI is redefining how people discover and buy online. The transformation brings uncertainty and excitement — but it’s not the first time technology has rewritten the rules of competition.
In 1969, that technology was the slide rule, a precision tool NASA used to calculate orbital paths for Apollo 11. Yet within a few short years, Hewlett-Packard’s HP 35 calculator made the slide rule obsolete. The HP 35 didn’t win by outperforming the slide rule at its own game. It changed the game entirely.
Political scientists call this a heresthetical shift, when a product, technology, or approach fundamentally redefines the competitive field instead of competing with it. HP’s calculator won by redefining the terms of advantage.
Two decades ago, Amazon changed what a bookstore could be and then transformed how people shop. Today, commerce faces its own heresthetical moment, driven by agentic AI which is fundamentally changing how people discover, evaluate, and buy.
When Brands, Retailers, and Makers Switch Seats
For years, commerce could be plotted on a simple 2×2 matrix based on what you sell and who you sell to. One axis represented single-brand versus multi-brand offerings, while the other distinguished between consumers and businesses. This model yields four distinct categories: brands, manufacturers, retailers, and distributors.
Now, those lines are blurring:
- Brands expand assortments and iterate faster, taking share from retailers.
- Manufacturers build direct relationships with individuals and businesses.
- Retailers create private labels and optimize supply chains and logistics, acting more like brands.
- Distributors expand value-added services and sometimes sell directly to consumers.
Examples abound. Lululemon thrives on tight brand control and owned distribution. Nike’s strong direct-to-consumer strategy reached over $5 billion in sales in late 2023, but is now recalibrating by strengthening its partnerships with Foot Locker and Macy’s.
Meanwhile, private label retail is surging, with merchants like Target and Walmart building brand equity with their own store brand lines. They are shifting from merely offering shelf space to brands, allocating space to their own labels, turning retailers into product creators rather than brokers.
What makes these changes heresthetical is how they redraw the boundaries of commerce. Companies across the retail ecosystem use data, experience, and AI to unify manufacturing, merchandising, distributing, and sales.
Enter Agentic Ecommerce
Agentic AI acts independently to achieve goals. Instead of following fixed instructions or producing one-off responses, it sets objectives, makes decisions, and takes action. In ecommerce, agentic systems operate across the funnel to drive measurable business outcomes.
There are three kinds of ecommerce agents, each with a distinct job:
- Agentic Clients: Software agents, like Perplexity, that shop on behalf of users.
- Agentic Services: Agents inside platforms like Coveo that interact directly with users on websites, and eventually with client agents, to retrieve relevant content, products, and answers.
- Agentic Merchants: Agents that work for merchandisers, optimizing product discovery, enriching catalogs, judging relevance, and generating targeted SEO landing pages.
There are three guiding principles to agents which include:
- Dedicated: Agents are singularly focused on specific tasks or skills that combine into broader workflows.
- Dialogical: LLM-powered agents with defined roles and perspectives that collaborate with humans and each other.
- Democratic: Decisions made collectively, with agents comparing viewpoints and voting to complete complex tasks
Agentic Merchandising in Practice
To illustrate how agentic AI changes merchandising, consider the kinetic sculpture Can’t Help Myself by Sun Yuan and Peng Yu. It featured a robotic arm endlessly scraping leaking hydraulic fluid toward itself—an image of futile reactivity.
Traditional merchandising often feels similar; caught up in reactivity. Teams constantly try to fix broken experiences with synonyms, redirects or rules. Agentic AI is the opposite approach. It’s proactive and anticipates problems and automatically corrects them at the source.
Agentic systems augment merchandisers so they spend less time patching systems and more time advancing strategy. Coveo supports three practical agentic skills that target root causes:
Catalog Enrichment
Agents inspect product photography, related content, and attributes to enrich missing metadata, the material that actually drives findability. Rather than hand-tagging or relying on static feeds, agents continuously improve data quality so search, browse, and recommendations work as intended. This provides more durable relevance.
Relevancy Judges
Agents identify relevancy tuning opportunities by evaluating whether results are relevant for a given query and flagging what’s off. They group issues into themes a merchandiser can address systematically. This automatically catches issues before they become executive big problems, resulting in fewer ad hoc fixes and more effective tuning.
SEO Boosters
Agents select keywords a site should rank for and omit those it shouldn’t. They generate precise, landing pages using semantic retrieval and agent voting, assembling the right products and excluding the wrong ones. This leads to more qualified organic traffic that converts.
Agents and merchandisers work together. Humans focus on storytelling, assortment, and experience design while agents handle behind-the-scenes optimization.
The Intent Box Revolution
For years, the search box has been the workhorse of ecommerce: type a keyword, hit enter, and get a grid of products ranked by relevance. This works great as long as you know what you’re looking for and can name it precisely. It struggles when language is imperfect, when you’re exploring, or when you need guidance before you’re ready to buy.
The intent box is the evolution of the standard keyword-to-results page model. It interprets what the shopper is trying to accomplish and adapts the experience accordingly. Sometimes that means returning a precise set of products with filters; other times, it offers concise guidance with links to categories, content, and curated items to buy.
For example, a search for “duck dummy” on a sporting goods website, likely means “duck decoy.” Even though the word “decoy” isn’t used in the query, the intent box interprets what the user wants, using a variety of AI capabilities including:
- Retrieval augmented generation (RAG)
- Vector retrieval
- Cosine similarity
- Named entity recognition
- Semantic understanding
- Lexical fuzzy matching
The familiar grid of results, featuring a variety of duck decoys, is maintained along with useful refinements like species, size, and motion. It corrects the language without adding friction.
For an educational query like “What do I need for duck hunting?” the system recognizes the goal,not a single SKU. The intent box responds with a concise explanation of essentials, links to relevant categories, articles and videos, and a shoppable set of products. Education and merchandising happen in one flow.
This adaptive behavior depends on detecting query intent and responding accordingly. Under the hood, four capabilities power this shift:
- Retrieval-augmented generation (RAG): Generates advisory answers grounded in the retailer’s own catalog and content, with citations. This avoids free-floating chatbot text and keeps recommendations on-brand and in-stock.
- Semantic understanding and vector retrieval: Understands meaning beyond exact words so phrases like “duck dummy” resolve to decoys. Uses cosine similarity to pull relevant products and content.
- Named entity recognition (NER): Identifies product types, brands, attributes, and relationships, aligning natural language with the language of the catalog.
- Lexical (including fuzzy) search: Delivers exactness on details that matter in commerce like size, compatibility, model numbers, and finish. This keeps results precise and shoppable.
Without dedicated, dialogical, and democratic agents focused on relevance, AI-powered commerce can produce bogus results. At Coveo, we believe relevance is the unsung hero, just like the slide rule was for the Apollo missions. Our unified architecture supports hybrid search, semantic understanding, and intent detection. This is what allows us to create an intent box that generates fluent answers and provides truly relevant, contextual shopping experiences that adapt to what users actually mean.
Why This Matters Now
This shift matters because it improves outcomes for everyone.
For shoppers, intent-aware experiences build confidence and trust by grounding answers in real products and content. The experience is precise when they know what they want and advisory when they don’t.
For merchandisers, AI agents take on the tedious work of enriching catalogs, judging relevance, and generating long-tail landing pages. This improves discoverability and opens new SEO avenues without cannibalizing core categories.
For businesses, fewer dead ends and more relevant paths drive higher conversions and AOV. Consistent, helpful guidance earns loyalty and creates a compounding advantage that is hard to copy.
The bigger point is heresthetical change. Calculators didn’t make slide rules better; they made them obsolete. Agentic AI is doing the same for ecommerce — replacing brittle rules with systems that understand intent, reason across products and content, and act to turn discovery into purchase.