Search is becoming conversational—but replacing a search box with a chatbot doesn’t automatically create a better shopping experience.
At Coveo, we’ve developed Conversational Product Discovery around a different idea: search itself should become conversational. Rather than forcing shoppers to choose between search and a separate chatbot, a single intelligence understands intent, retrieves the right information, and renders the most appropriate experience. One intent, one query, one moment—not two competing interfaces bolted together.
That’s the vision. Delivering on it, however, requires solving four challenges at once: understanding intent, grounding responses in trusted product data, responding fast enough to keep shoppers engaged, and presenting answers in the right interface. That last piece is where a protocol called AG-UI becomes essential, and we’ll get to why.
Challenge 1: Understanding intent
Going conversational means handling a much wider range of intents than a traditional search bar ever had to.
Shoppers don’t just search for products anymore. They compare items, ask for bundles, refine results through conversation, request recommendations, explore alternatives, or expect results to reorganize themselves as their needs become clearer. That’s a fundamentally richer job than matching keywords to a product list.
Many early conversational shopping experiences have exposed the challenge. Some assistants recommend products effectively but struggle with everyday shopping tasks like comparing products, refining results, or answering specification questions. Those aren’t edge cases—they’re central to how people shop. If a conversational experience can’t support them, it isn’t replacing search with something better, it’s replacing it with something narrower.
Challenge 2: Grounding
Understanding intent is only half the problem. The answer you provide also needs to be correct.
When shoppers ask whether two products differ, whether a component is compatible, or whether an item is in stock, responses must be grounded in trusted catalog data — not generated from the model’s general knowledge. Hallucinated product specifications or outdated inventory make for poor experiences, undermining trust and can directly impact revenue.
Conversational Product Discovery therefore depends on retrieval as much as generation. Large language models excel at reasoning and conversation, but they need accurate, up-to-date product information to reason over.
Challenge 3: Speed
Even with the right intent and grounded information, conversational experiences live or die on responsiveness.
Shoppers accustomed to instant search results won’t wait several seconds for an answer. Time-to-first-token, streaming, intelligent caching, and efficient retrieval all become essential. Significant progress is being made here, but latency remains a design constraint every conversational commerce experience has to respect.
This is also where interface architecture quietly becomes a speed problem. Generating a bespoke front end for every response means waiting on a model to produce and validate new code before anything can render. Rendering from a library of components that are already built, tested, and cached skips that step entirely. The model only has to decide which component to use and hand it data. That difference in architecture is often what separates a response that feels instant from one that feels like a demo.
Challenge 4: Adaptive UI
This is the challenge that perhaps receives the least attention, and arguably the one that determines whether conversational shopping actually feels better.
Different intents naturally call for different interfaces. A shopper comparing two products expects a comparison table, not a paragraph. Someone browsing outfit ideas expects visual product cards. Someone refining a search expects filters that can be adjusted interactively. Someone asking about battery life wants a concise, grounded answer, and a shopper looking for complementary products expects a curated carousel or bundle, not several back-and-forth messages to get there.
Trying to force every one of these experiences into a single chat bubble is like trying to display an entire ecommerce site inside a messaging app. It works for some tasks, but quickly becomes limiting.
The goal isn’t to answer every shopping question inside a conversation. The goal is to let the interface itself adapt to the shopper’s intent.
This is where AG-UI comes in.
So What Is AG-UI?
It’s worth being precise here, because the term gets used loosely.
AG-UI, short for the Agent-User Interaction Protocol, is an open standard, originated by CopilotKit, that defines how an AI agent communicates with a front-end application in real time: what it’s doing, what’s changed, and what the shopper should see next. It’s the transport layer, not a UI style in itself.
What matters for shoppers is the pattern Coveo builds on top of that protocol. Generative UI, broadly, covers a spectrum. At one end, an agent can generate an entire interface from scratch for a single interaction. That’s genuinely impressive, and well suited to open-ended, low-volume workflows, but expensive to run and hard to govern once you’re serving millions of broadly similar shopping journeys a day, each needing predictable performance, accessibility, and brand consistency.
Coveo’s Conversational Product Discovery sits at the other end of that spectrum. The agent doesn’t generate front-end code. It selects which predefined, production-tested component (comparison table, product carousel, filter panel, spec sheet, recommendation module) best answers the shopper’s intent, using AG-UI to stream that decision and the underlying data to the front end. The component itself was already built, tested, and approved; the agent is choosing, not authoring.
Put differently:
Open-ended generative UI asks: What could the model build for this interaction?
Coveo’s approach, running on AG-UI, asks: Which proven component best solves this shopping task?
This distinction is what delivers much of the responsiveness shoppers associate with a conversational interface while preserving the consistency, speed, governance, accessibility, and operational efficiency enterprise retailers require.
Bringing It Together
Intent understanding, grounding, speed, and adaptive UI aren’t independent capabilities, they’re the foundation of conversational product discovery.
Get intent wrong, and shoppers receive irrelevant answers. Get grounding wrong, and they lose trust. Get speed wrong, and they abandon the experience. Get the interface wrong, and even correct answers become frustrating to use.
AG-UI makes that final piece practical. Rather than forcing every interaction into a chat window, it allows the shopping experience itself to adapt to each intent while remaining fast, governed, and on-brand.
Conversational Product Discovery is about giving shoppers the right interaction for the intent they’re expressing. Sometimes that’s a conversation. Sometimes a comparison table, a product grid, or a concise answer grounded in trusted product data.
The intelligence remains the same. The interface adapts.