A year ago, “AI agents will do your shopping” was a keynote slide. It is now a traffic source.
AI-referred traffic to US retail sites grew 393% year over year in the first quarter of 2026, according to Adobe Analytics. Over the 2025 holiday season, Salesforce measured AI and agents touching roughly 20% of retail sales, around $262 billion in spending. McKinsey projects $3 to $5 trillion in global agent-mediated commerce by 2030, and Bain expects 15% to 25% of US e-commerce to run through agentic channels by the same year.
Read the range rather than any single number. Different firms define “agentic” differently. But every forecast points the same direction, and the current data already backs it up.
So the strategic question for brands is no longer whether agents matter. It is narrower and more useful: what does an agent do well, what does it structurally fail at, and what should you build to cover the gap?
What an agent is good at, and what it cannot do
An AI agent is very good at narrowing. Give it a budget, a constraint, and a deadline, and it will search across retailers, compare specs, check reviews, verify availability, and hand back a short list. That is genuine work, done faster than a person would do it.
What an agent cannot do is make you believe.
There is a category of purchase where the last step is not evaluation, it is confidence. Will this shade suit my skin. Will this chair fit that corner. Will this frame look ridiculous on my face. An agent can tell you the answer is yes, with sources. It cannot show you.
That gap has a measurable cost. Mirakl found that 42% of customers abandon purchases because there is not enough product information. On the back end, US retail returns reached nearly $850 billion in 2025 according to the National Retail Federation, and a large share of those in apparel and beauty come down to the product not matching what the shopper pictured.
Agents make the front of the funnel faster. They do nothing about that.
AR is the confirmation layer
This is the useful way to think about augmented reality in an agentic world. Not as a competing discovery channel, and not as a marketing add-on. As the step between “the agent recommended this” and “I believe it.”
The pattern looks like this. The agent does the research and narrows to two or three options. The shopper opens the product page. Instead of a static gallery, they see the thing on themselves or in their space, in their lighting, in real time. Then they buy, or they do not, but either way they decide with information rather than hope.
The evidence for that final step is reasonably good. Shopify’s Gunner Kennels case reported a 40% increase in order conversion and a 5% reduction in returns after adding a life-size 3D model that customers could place next to their own dog. A 2025 peer-reviewed study on AR cosmetic try-on, cited by Shopify, found that it improves a shopper’s ability to evaluate a product and makes the purchase feel less risky.
Less risky is the operative phrase. That is exactly the job an agent leaves undone.
We built noyah’s AR try-on around this. Clean beauty, real shade range, and the oldest problem in online cosmetics: a shopper who cannot tell how a color will read on their own skin. Face tracking and color accuracy do not make the recommendation. They confirm it.
The problem nobody is budgeting for yet
Here is the part that should worry brand teams more than it currently does.
When an agent handles discovery, the brand loses the behavioral data that used to come with it. The browsing, the comparison, the hesitation, the refinement of what someone actually wants: all of that now happens inside the agent. What reaches the merchant is the add-to-cart event and nothing before it.
Attribution gets harder. Personalization gets thinner. The feedback loop that made your last three campaigns smarter starts to close.
AR is one of the few surfaces where that data still belongs to you. When someone opens a try-on or scans a package, every interaction is a first-party signal given voluntarily:
- Which shades or variants get compared against each other
- How long someone stays with a product before deciding
- Which features people zoom in on
- The exact point where they hesitate and leave
None of that is inferred or bought. It is observed, on your property, by your experience. In a market where the discovery layer is increasingly rented from someone else, an owned interaction surface is worth more than it was two years ago, and not only for the conversion it produces. If you want the framework for measuring it, we wrote about the KPIs that actually matter in an XR campaign.
What to build, in order
Three principles, in the order they matter.
Put the confirmation where the hesitation is. Not on the homepage, not in a campaign microsite. On the product page, on the packaging, at the exact moment someone is deciding. We wrote about the packaging version of this separately, because a box a shopper is already holding is the highest-intent surface most brands own and the one they use least.
Make it openable in one step. If confirmation requires a download, it is not confirmation, it is a second decision. WebAR runs in the mobile browser from a QR code or a link, which is the only version of this that survives contact with a real shopper. An agent that hands off to an app store install has handed off to nothing.
Instrument it from day one. The data described above only exists if you set it up to be captured. Retrofitting analytics onto a live AR experience is possible and annoying. Building it in costs nothing extra.
Where this actually lands
The loud version of the AI and AR story is that they merge into one interface: an assistant living in your glasses, narrating the world. That may happen. It is not a 2026 procurement decision.
The quiet version is more useful and available now. Agents are taking over the part of shopping that is research. That leaves brands competing on the part that is conviction. AR is the most direct tool anyone has for producing conviction at the moment of decision, and it runs on the phone every one of your customers is already holding.
The brands that treat AR as a novelty budget line will keep getting novelty results. The ones that treat it as the confirmation layer under an increasingly automated funnel will be the ones agents keep recommending, because their products come back less and convert more.
Frequently asked questions
What is agentic commerce? Agentic commerce is shopping carried out by an AI agent on a person’s behalf. Rather than browsing and comparing manually, the shopper states an intent and the agent researches options, compares them, and in some cases completes the transaction.
Does AR compete with AI shopping agents? No. They solve different problems. Agents narrow a large set of options down to a few. AR resolves the doubt that remains about the ones that are left. A brand that has one and not the other has an incomplete funnel.
Does AR still work if a customer arrives from an AI agent? Yes, and it matters more in that case. A shopper arriving on an agent’s recommendation has skipped their own research. The confirmation step is the only interaction they have with your brand before deciding.
What kind of AR should a brand build first? Whichever one sits closest to the point of hesitation in your category. Beauty and eyewear start with try-on. CPG usually starts with packaging. Furniture and hardware start with placement. Build the one that answers the question your customers actually ask before buying.
Do customers need to download an app? Not for WebAR. It runs in the mobile browser from a QR code, a link, or a product page. That matters, because a download requirement is enough friction to lose most of the traffic it is meant to convert.
What can a brand actually measure from an AR experience? Opens, dwell time, variant or shade comparisons, feature-level interaction, drop-off points, and add-to-cart behavior for users who engaged versus those who did not. All of it is first-party.
Building the confirmation layer
If shoppers are hesitating somewhere in your funnel, that hesitation is the business case. We build utility-first AR for CPG, retail, and enterprise, and it runs on the phone your customer already has.
Sign up for a Virtual Fudge Demo 30 min or 15 min intro call
Virtual Fudge is a New York City augmented reality studio. See our work.