What Salesforce shipped
Salesforce has made Agentforce Commerce generally available, calling it the biggest commerce release the company has shipped in five years. The package puts three purpose-built AI agents into its commerce cloud: a Shopper Agent for consumer storefronts, a Buyer Agent for B2B ordering, and a Merchant Agent for back-office merchandising. All three are live now rather than in preview. The pitch to retailers is that generative shopping is arriving whether they are ready or not, and that the safest place to meet it is on properties the brand already controls. Nitin Mangtani, executive vice president and general manager of Agentforce Commerce, has argued that commerce will happen predominantly on owned and operated properties even as discovery shifts to AI platforms.
The timing is deliberate. Salesforce shipped ahead of the peak holiday season, giving retailers a runway to deploy before the demand spike that determines the retail year. The company cites its own data that AI influenced roughly 20 percent of global online sales last holiday season, about 262 billion dollars, and that retailers running shopper agents grew sales 59 percent faster than those without. Whether those figures hold up under scrutiny matters less than the direction they signal. Salesforce is betting that conversational, agent-mediated shopping moves from novelty to baseline within a year, and it wants its install base of commerce customers to adopt agents through Salesforce rather than through a rival or a raw model API.
Three agents, three surfaces
The Shopper Agent is the consumer-facing piece. It handles discovery, product advice, checkout, and post-purchase service directly on a brand's storefront, aiming to replace the static search-and-filter experience with a conversation. The Buyer Agent targets B2B, running procurement through WhatsApp and SMS with multi-cart and quoting capabilities, which fits the reorder-heavy workflows of wholesale and distribution. The Merchant Agent turns back-office tasks into natural-language commands, letting merchandisers organize catalogs and adjust merchandising without clicking through admin screens. The three-agent split maps cleanly onto the three constituencies of any commerce operation: the end shopper, the business buyer, and the internal operator. Each gets an agent tuned to its context rather than one generic assistant stretched across all of them.
That segmentation is the interesting design choice. Many vendors are shipping a single shopping assistant and calling it agentic commerce. Salesforce is instead betting that the buyer and the merchant have needs distinct enough to warrant separate agents with separate permissions and data access. The Buyer Agent living in WhatsApp and SMS is a particular tell: B2B ordering in much of the world already happens over messaging, and meeting buyers there rather than forcing them onto a portal reflects how the work actually gets done. Shirley Gao, chief digital and information officer at PacSun, framed the consumer side as an opportunity to extend products into AI platforms like ChatGPT. The architecture assumes agents proliferate across surfaces the brand does not own.
The bet on owned storefronts
Mangtani's thesis is worth taking seriously because it cuts against the loudest narrative in agentic commerce. The prevailing story says shopping migrates into ChatGPT, Gemini, and Perplexity, and that merchants become anonymous suppliers to whichever AI intermediates the transaction. Salesforce is arguing the opposite for the moment that matters. Discovery and referral will be shaped by AI platforms, yet the actual purchase, the checkout, the account, and the service relationship will concentrate on the brand's owned storefront. If that holds, the storefront remains the strategic asset and the agent running it is the control point. Salesforce, which sells storefront software, has an obvious commercial interest in that outcome, so the claim deserves scrutiny alongside its logic.
There is a credible case for it. Checkout involves payment credentials, loyalty status, warranty, returns, and personalized pricing, all of which live in the merchant's systems and are awkward to fully delegate to a third-party agent. A brand also wants the post-sale relationship, because that is where repeat revenue and margin sit. The risk to Salesforce's thesis is that consumers simply prefer transacting inside the AI assistant they already trust, and that payment rails like the emerging agentic commerce protocols make third-party checkout frictionless enough to win. The honest answer is that nobody knows yet which surface captures the transaction. Salesforce is hedging by connecting its storefront agents to the external platforms rather than pretending they will not matter.
The ChatGPT and Google connection
The integrations are the practical hook. Salesforce says the Shopper Agent connects natively to ChatGPT, with that integration reaching general availability in July 2026, and to Google Search including AI Mode and the Gemini app arriving over the summer. Mechanically, a retailer's product catalog syncs from Salesforce's business manager straight into ChatGPT with no additional software or third-party middleware. That lets a brand appear inside the AI assistants where discovery is migrating while still routing fulfillment and service through its Salesforce storefront. It is a bridge strategy: participate in the external agent ecosystem for reach, retain the owned storefront for the transaction and the relationship. For retailers already on Salesforce, the marginal effort to light this up is low.
The competitive context is crowded. BigCommerce is shipping agents that build carts and generate checkout links over the Model Context Protocol, and every major commerce platform is racing to connect merchant catalogs to ChatGPT, Perplexity, and Copilot. Google's Universal Commerce Protocol and the OpenAI and Stripe Agentic Commerce Protocol are competing to become the standard rails underneath all of it. Salesforce's advantage is distribution: it already runs commerce for a large base of enterprise retailers, so shipping agents into that base is faster than a challenger winning those accounts from scratch. The risk is that platform-native agents from OpenAI or Google commoditize the storefront layer. Salesforce is defending the high-value part of the stack while it still owns the customer.
Where the numbers come from
Enterprise buyers should read the supporting metrics carefully. The claim that retailers with shopper agents grew sales 59 percent faster, and that AI-referred traffic converts eight times higher than social, comes from Salesforce and its data partners rather than from an independent audit. Vendor-supplied performance figures reliably describe early adopters, who tend to be better-resourced and more digitally mature than the median retailer, which inflates the apparent lift. That does not make the numbers useless. It makes them a ceiling rather than an expectation. A CIO evaluating Agentforce Commerce should ask what the comparable adopters looked like before deployment and model a fraction of the headline gain rather than the full figure.
The more durable signal is the 20 percent of online sales that AI influenced last holiday, which aligns with what other analysts and platforms are reporting independently. AI-mediated discovery is clearly material and growing, and the strategic question for retailers is how to be present in it without ceding the customer. That is the problem Agentforce Commerce is built to solve, and on that framing it is well positioned regardless of whether any single conversion statistic survives scrutiny. The practical test for buyers is integration effort against measured incremental revenue in their own environment. Run a bounded pilot, instrument it honestly, and let the retailer's own data settle whether the agents earn their keep.
Our read
Agentforce Commerce is a confident, coherent product, and its underlying bet is the most defensible position we have seen a large vendor stake in agentic commerce. The argument that discovery fragments across AI platforms while the transaction concentrates on owned storefronts is plausible, commercially self-serving, and testable. Salesforce has hedged it well by connecting its agents to ChatGPT and Google rather than walling them off. For the enterprise-scale retailers already running Salesforce commerce, the release lowers the cost of experimenting with agents to something close to a configuration exercise, which is exactly how an incumbent should defend an install base against faster-moving challengers.
The caution is dependency. Adopting Salesforce's agents deepens reliance on a single vendor for storefront, agent orchestration, and the connective tissue to external AI platforms. That is convenient now and a lock-in risk later, particularly if pricing on agent usage climbs once the capability becomes load-bearing. Retailers should adopt with eyes open, keep their catalog and customer data portable, and avoid architecting so tightly around Salesforce agents that switching becomes impractical. The strategic prize, staying present in AI-mediated shopping while owning the transaction, is real and worth pursuing. The discipline is to pursue it without handing any one vendor the entire commerce stack. That balance, more than any conversion statistic, will separate the winners.
