What Is Agentic Commerce?
- 4 days ago
- 11 min read
 A note from Lauren, before you dig inFOUNDER'S NOTE I wanted to open this guide with something that's been on my mind longer than agentic commerce has been a marketing buzzword. Every few years, a new way to shop shows up, and every time, a wave of predictions follows about how it's going to rewrite the rules of how people choose what to buy. It happened when ecommerce arrived. It happened again with mobile. It's happening now with AI agents doing some of the shopping for us. And every single time, the fundamentals of how a brand actually grows have held up better than the hype cycle wanted to admit. Here's the thing that's stayed true through all of it: people buy brands they're familiar with. That's not a hot take, it's one of the most well established findings in marketing, but it's worth restating because it's easy to lose sight of in a room full of people talking about algorithms and agents. Familiarity gets built the same way it always has, through advertising, PR, and the kind of broad-reach marketing that gets a brand in front of people long before they're actually in the market to buy anything. It is not built through a perfectly tagged product feed or a clever prompt strategy. Those things matter once someone is already shopping. They do very little to make someone want to shop for you in the first place. That distinction matters a lot right now, because a lot of brands spent the last several years quietly starving their brand budgets in favor of performance media, chasing efficient, measurable, bottom of the funnel wins. The chickens have come home to roost. Agentic commerce is just the newest, shiniest reason to keep making that same mistake: pouring money into paid placement inside an AI tool and calling it brand strategy. It isn't. It's a lower funnel tactic wearing a new outfit, and the early research backs this up. A brand's relative positioning inside AI tools is itself shaped by existing brand equity and reputation. Real world brand building feeds into how AI tools treat you. It does not work the other way around. So consider this guide two things at once. It's a genuinely useful primer on what agentic commerce is, how it actually works today, and where the credible research says it's headed, because it is a real shift worth understanding. And it's my reminder that for most CPG brands, heavy investment in agent-specific tactics right now is premature. The juice isn't worth the squeeze yet. Keep building awareness, consideration, and purchase intent the way brands always have. That's still what gets you remembered and picked, whether the one doing the picking is a person or a bot trained on how people already behave. This is also a living document. Agentic commerce is still genuinely early, and we'll keep updating this guide as the space matures and the research catches up to the hype. — Lauren Ridgley CEO, Left Hand Agency |
What is agentic commerce?
Agentic commerce is the shift from AI helping you shop to AI actually shopping for you.
Assisted search, like asking ChatGPT or Google to summarize product reviews, isn't new anymore. Agentic commerce goes a step further: an AI agent researches, compares, adds to cart, and in some cases completes checkout, sometimes without a person approving that specific purchase.
Bain & Company draws a useful line here, defining agentic commerce as transactions “initiated, influenced, or completed by third-party AI agents or retailer-hosted agents,” which specifically excludes basic AI-assisted search from the definition.
The infrastructure for this caught up fast. Walmart launched its shopping assistant Sparky in June 2025. OpenAI rolled out Instant Checkout inside ChatGPT that September, built on an open standard called the Agentic Commerce Protocol, developed with Stripe. Perplexity followed with “Buy with Pro” that November. On the payments side, Google published its Agent Payments Protocol with more than 60 partners, including Mastercard, American Express, and PayPal, and Mastercard has since extended into agent-to-agent payments for machine-to-machine transactions. The pieces all exist now. What's still an open question is how much of the process consumers actually want to hand over.
The spectrum: from “help me shop” to “just buy it”
It helps to think of agentic commerce as a range rather than one thing.
At the assisted end, AI narrows down choices but a person still clicks buy. This is where most current usage sits. Walmart's Sparky synthesizes reviews and answers comparison questions, but Walmart frames it as taking “the guesswork out of shopping” so a shopper can “add to cart with confidence,” not as something that adds to cart on its own.
In the middle is assisted shopping with checkout built in, like ChatGPT's Instant Checkout or Perplexity's Buy with Pro, where a purchase happens inside the chat window itself. Worth noting: OpenAI actually walked part of this back in March 2026, shifting toward letting retailers like Target, DoorDash, and Instacart handle checkout inside their own apps rather than inside ChatGPT directly. Shopify's president pointed out why: checkout isn't just payment, it's subscriptions, inventory, shipping, and tax rules that retailers don't want to hand to a third party. That's an early signal that “buy it inside the chat” is harder to execute cleanly than the initial pitch made it sound.
At the far end is delegated or fully autonomous purchasing, where the AI acts without a human confirming that specific transaction. Amazon's Alexa for Shopping (formerly Rufus) now has an auto-buy feature for Prime members that completes a purchase using saved payment info, with a notification and a 24-hour cancellation window instead of an upfront approval. Google's AP2 protocol has a formal mechanism for this, a “delegated task,” where a user pre-authorizes a condition (“buy these tickets when they go on sale”) and the agent executes later without asking again.
Where it's actually headed
Every major forecaster now has a number on this. McKinsey projects agentic commerce could reach $3 to $5 trillion globally by 2030. Bain estimates $300 to $500 billion in the U.S., or 15 to 25% of all U.S. ecommerce, by the same year. Morgan Stanley lands a bit lower, at $190 to $385 billion, or 10 to 20% of U.S. online retail. eMarketer's own modeling is in a similar range.
Two patterns show up across almost all of this research and matter more than the topline dollar figures.
First, adoption won't be even across categories. Commodity, replenishment-style purchases like batteries, paper towels, and household basics will convert to agentic buying fastest, because the decision is about price and convenience, not identity or taste. More considered purchases will lag.
Second, Morgan Stanley specifically flags groceries and CPG as the category already leading AI-driven purchases today, and the one expected to see the largest growth over the next five years. That's directly relevant if you're in this industry: the runway here is shorter than in some other retail categories.
There's also a metric worth watching for. Oliver Feldwick, VML's Chief Innovation Officer, predicts that “share of model” will emerge as a meaningful brand health metric over the next couple of years, the AI-era equivalent of “share of search.” He expects it to sit alongside traditional mental availability metrics with humans rather than replace them, since a CPG brand still has to win the category entry point in someone's head regardless of whether a person or a model is doing the choosing. Translation: the old job doesn't go away, it just gets a second scoreboard.
Signal vs. noise: separating what's real from what's hype
Here's the part that doesn't get said enough. A lot of the current conversation around agentic commerce is running ahead of the evidence, and some credible analysts are actively pushing back.
Gartner's 2026 hype cycle places agentic AI squarely at the “Peak of Inflated Expectations,” the stage right before reality sets in and expectations correct downward. Their data shows over 60% of organizations expect to deploy AI agents within two years, but only 17% actually have so far. That's a wide gap between stated intent and what's actually built.
Forrester's own 2026 predictions come with a built-in warning label. They expect a third of retail marketplace projects to fail as brands chase answer engines, while noting only 24% of U.S. adults currently use ChatGPT at all, meaning some of that shift looks more like panic than proven demand. Forrester also predicts a fifth of B2B sellers will face some form of agent-led negotiation, but adds a direct caveat: “true autonomy and broad use are still in the distance.” Their framing for some of these bets is blunt: “learning experiences at best and money pits at worst.”
Retail media analyst Andrew Lipsman went further, calling some of the industry discourse a “collective hallucination,” arguing that dramatic projections of margin erosion and advertising revenue collapse from AI agents don't hold up under scrutiny, and that the intensity of the conversation reflects industry anxiety (especially around struggling retail media networks) more than measurable consumer behavior.
The actual usage numbers back up some of that skepticism. Alipay's AI shopping agent processed over 120 million transactions in a single week in China in February 2026, which sounds enormous, but only around 8% of Western consumers have ever let an AI agent complete a purchase for them. Agentic commerce is genuinely moving fast in specific corners (retailer-owned assistants, payment infrastructure, B2B) while still being nowhere close to normal behavior for the average American shopper.
The honest read: the infrastructure is real and being built quickly. The idea that most people will hand over full purchasing control to a bot in the next year or two is not. |
Does brand still matter?
This is the reassuring part, with one important nuance. Research consistently shows AI is changing how people discover products, not how they ultimately decide.
On discovery, AI is opening real doors. eMarketer research found that 51.7% of consumers said generative AI had introduced them to a brand they didn't previously know, and described that experience positively. But when it comes to whether an AI includes a brand in that shortlist in the first place, brand size matters less than you'd think. One study testing how AI models choose which brands to recommend found company size explained only 22% of the outcome, while documentation quality and third-party proof pushed that number to 41%.
In broad, generic searches, smaller challenger brands got about 14% of recommendations. In specific, high-intent searches, that jumped to 38%. In other words, a well-documented smaller brand has a real shot at landing next to a household name when the search is specific enough, which is genuinely good news for challenger CPG brands.
Where established brand equity reasserts itself is at the final decision. Research cited by PSE Consulting found that 89% of shoppers say recognizing the seller's brand is important or very important even after AI has narrowed the field, and 92% still check reviews before choosing between AI-suggested options. Only 14% simply take whatever the AI ranks first. Separately, only 6.3% of consumers said they'd let an AI suggestion override an established brand preference entirely.
AI builds the list. Brand equity still wins the pick.
There's one more wrinkle worth flagging: paid placement seems to backfire here in a way it doesn't in search or social. Research from The Harris Poll and Quad found that 75% of consumers said they'd trust a brand less if they found out it paid to influence an AI's recommendation. People currently expect these answers to feel neutral, and that expectation is fragile.
Agentic Commerce Guide: What CPG brands should actually do
Given everything above, both the research and the reminder up top about how brand growth actually works, here's how I'd split your priorities: what's worth doing now, and what can genuinely wait.
DO THIS NOW
Get your product data clean and complete. Ingredients, allergens, certifications, sizes, and claims should be structured and consistent everywhere your product appears. This is the single biggest factor in whether an AI agent surfaces you at all, and it matters more than brand size.
Build real third-party proof. Reviews on your own site, on retailer pages, and in places like Reddit all feed what AI agents lean on to validate a product. Agents increasingly trust what other people say over what a brand says about itself.
Make your listings consistent across every channel. An agent might pull information from your D2C site, a retailer page, or a third-party database. Inconsistent claims or pricing across those sources creates the kind of doubt that gets a product dropped from a shortlist.
Earn visibility instead of buying it. Given how strongly consumers react against paid influence inside AI answers, invest in the things that create organic visibility (data quality, reviews, genuine differentiation) rather than trying to pay your way into a recommendation.
Ask AI models what they say about you. Take the category questions your customers already ask, something like “what should I look for in a gluten-free bread mix” or “best electrolyte powder for kids,” and run them across a few different AI models and a few different phrasings. See whether your brand shows up, and whether the traits it's known for actually come through. Oliver Feldwick, VML's Chief Innovation Officer, has pointed out that small wording changes (like “premium” versus “luxury”) can shift the results meaningfully, so test more than one version of the question rather than assuming one try tells you anything. It's a free, fast way to see the gap between how you think your brand is perceived and how the models currently describe it, and it tends to be a genuine wake-up call.
OPTIMIZING FOR THE RETAILER ASSISTANTS THAT ALREADY EXIST
Walmart's Sparky and Amazon's Alexa for Shopping are live today and already driving real orders, and this is where the near-term opportunity actually is, not in some future brand-owned AI agent. “Optimize” means something specific and different for each one.
Amazon's Alexa for Shopping runs on Amazon's COSMO system, which reads far more of your listing than a typical shopper does. It pulls from backend search attributes in Seller Central (fill in every field, not just the required ones), product bullets, the full 2,000-character description, A+ content like comparison charts and FAQ modules, your Q&A section, and customer reviews, which the system mines for specific use cases and recurring complaints. Amazon reportedly surfaces only around five products per query, so an incomplete listing does not lose out on ranking so much as it disappears from consideration entirely. Writing bullets as outcome, then feature, then use case (rather than a string of keywords) and seeding your own Q&A with real questions and detailed answers both make a measurable difference.
Walmart's Sparky works less like a keyword index and more like an intent matcher. Walmart's own guidance and third-party research both point to the same thing: Sparky synthesizes your product description to understand what a product is for, rather than scanning it for keywords, so complete catalog attributes (size, use occasion, material, ingredients) and description copy written the way a real customer talks about their need matter more than SEO-style phrasing. Reviews that describe a specific use case carry more weight in this system than a plain star rating.
One caveat: Walmart has started selling sponsored placement inside Sparky's answers, similar to a sponsored search ad. Given the brand trust research above on how consumers react to paid influence inside AI answers, treat that as a supplement to strong organic content, not a shortcut around it. |
WATCH AND WAIT
Building your own branded shopping agent or chatbot. For most CPG brands, this isn't where the action is. Consumers are using retailer- and platform-owned agents, not visiting individual brand sites for an AI concierge experience.
Big spend on unproven “AI visibility” agencies and tools. This space is growing fast and the measurement standards aren't settled yet the way they are for SEO. Some of what's being sold as guaranteed AI ranking is closer to hype than a proven discipline right now.
Full agent-to-agent negotiation and automated pricing systems. This is genuinely interesting long-term infrastructure, but it's largely a B2B and enterprise-level concern for now. Even Forrester's own prediction for this, a fifth of B2B sellers by an unspecified date, comes with the caveat that broad, true autonomy is still far off.
Machine-to-machine payment rails. Emerging tools built for background, agent-to-agent microtransactions are interesting to know about, but they aren't something a brand marketing team needs a strategy for today.
Abandoning traditional brand building. The data is clear that established brand equity and genuine customer reviews still decide the final purchase, even inside an AI-assisted shopping journey. Don't redirect your whole budget chasing AI visibility at the expense of the brand fundamentals that are still doing the heavy lifting.
We are Left Hand Agency, a CPG media buying agency helping brands grow with short and long-term strategies. Our memory-driven strategies deliver results your marketing and finance teams will champion.
