Agentic Commerce Crosses From Demo to General Availability
On July 6, 2026, Salesforce declared its biggest Agentforce Commerce release yet, moving three agents to general availability. The Shopper Agent runs conversational shopping on the storefront with live inventory awareness, the Buyer Agent handles B2B procurement over WhatsApp and SMS with multi-cart support, and the Merchant Agent lets operators run back-office tasks through natural language. The significance is the GA label. For two years agentic commerce lived in keynote demos and limited pilots, and technology leaders rightly discounted it. A vendor of Salesforce's scale putting production agents behind a supported release changes the risk calculus for anyone still treating AI agents as experimental.
The release lands as AI-driven shopping stops being hypothetical. Salesforce reports that AI influenced 20 percent of global online sales, about USD 262 billion, during the past holiday season. That is a large enough share that ignoring it becomes a strategic liability rather than a defensible wait-and-see posture. The company is betting that merchants will want agents embedded across the full journey, from a shopper asking questions on the site to a merchandiser adjusting a promotion by typing a sentence. For commerce leaders, the question shifts from whether agentic commerce is real to how quickly their stack can support it without a rebuild.
The Numbers That Justify the Investment
Salesforce came armed with conversion data, and the figures are the reason to pay attention. The company says AI-referred traffic converts at eight times the rate of social media traffic, a gap that reframes where marketing budget should flow. It also claims retailers running shopper agents grew sales 59 percent faster than competitors, and that Merchant Agent users cut time to task by 88 percent. Vendor-supplied numbers always warrant scrutiny, yet even discounted heavily they point in one direction: high-intent AI referrals monetize far better than the social traffic many merchants have overinvested in for a decade.
For a margin-conscious operator, the productivity figure may matter more than the conversion one. An 88 percent reduction in time to task inside the back office is a labor story, and labor is where retail margin quietly erodes. If a merchandiser can reprice, re-merchandise, or launch a campaign by describing intent instead of navigating menus, the effective cost of running the storefront drops. That efficiency is easier to underwrite than speculative topline lift, because it attacks a cost the business already carries. The disciplined way to pilot Agentforce Commerce is to instrument both the conversion claim and the operational one, and let the harder-to-fake savings anchor the case.
Discovery Leaves Your Site, Transactions Stay
The strategic frame Salesforce is pushing deserves its own analysis. EVP and General Manager Nitin Mangtani argued that referral traffic will be heavily influenced by AI platforms like Gemini and ChatGPT and by social apps such as Instagram and TikTok. His load-bearing claim is where the money lands: Mangtani said commerce itself will happen predominantly on owned and operated properties, your website, your app, your messaging channels, and your stores. That split, discovery on AI platforms and transaction on owned properties, is the mental model every commerce leader should adopt when deciding what to build and what to plug into.
If Mangtani is right, the implication is concrete. Merchants should treat AI platforms as a new top-of-funnel channel to be optimized and measured, the way they once treated search and social, while investing hard in owned properties as the place value is actually captured. That argues against ceding the transaction to a third party's walled garden, because the order record, the customer relationship, and the margin all concentrate at checkout. The teams that thrive will make their catalog and inventory legible to external agents for discovery while keeping the closing experience, and the data it generates, firmly under their own control.
ChatGPT Checkout Goes Live and Forces a Decision
The release makes native checkout inside ChatGPT generally available in July, with Google Search AI Mode and the Gemini app following over the summer. Products sync directly from Business Manager, and Salesforce is explicit that the merchant remains the order record holder. That detail is the whole ballgame. It means a brand can accept an order initiated inside ChatGPT while still owning the transaction data, the fulfillment relationship, and the post-purchase experience. The alternative, letting the AI platform intermediate the order and own the record, would strip the merchant of exactly the customer data that funds retention and personalization.
This puts a live decision on the table for commerce leaders. Enabling agent checkout expands reach into a channel Salesforce says converts eight times better than social, yet it also normalizes shoppers buying without ever visiting your storefront. The right posture is to participate on terms that preserve data ownership, then measure incrementality honestly. Some of these orders will be genuinely new demand, and some will cannibalize direct sales you would have captured anyway at higher margin. Instrument the difference before scaling. Reach into ChatGPT is only valuable if it grows the pie rather than rerouting existing customers through a channel that dilutes your economics.
The Platform Play Under the Agents
Beneath the agents sits infrastructure that reveals Salesforce's real strategy. The release unifies order management across what the company describes as 7,000 locations and 250 million records, ships a Storefront Next that promises 30-minute deployment, and brings a modern POS to iOS now with Android and Windows following in August 2026. The catalog is unified across point of sale and storefront. Named customers including PacSun, represented by Chief Digital and Information Officer Shirley Gao, and Iceland Foods, whose Head of Ecommerce Technology Luke Barber participated, lend the release operational credibility beyond the demo stage.
The unification is the point, and it is where build-versus-buy gets serious. Agents are only as good as the data and inventory they can act on, so a fragmented stack with separate systems for POS, ecommerce, and order management starves them. Salesforce is selling the integrated substrate as much as the agents themselves, which is a classic platform lock-in move. Leaders should weigh the genuine benefit of a coherent data layer against the cost of concentrating their commerce operations inside one vendor. The efficiency is real, and so is the dependency. Neither should be adopted without a clear-eyed view of the switching costs it creates.
How to Sequence Your Agentic Commerce Bet
The pragmatic reading of this release is that agentic commerce has reached the stage where waiting carries more risk than acting, but acting recklessly carries plenty too. The defensible sequence starts inside the walls. Deploy the Merchant Agent against back-office tasks where the 88 percent time savings is measurable and the customer never sees a failure, then expand to customer-facing shopper agents once the team trusts the guardrails. Internal automation builds the operational confidence and the data discipline that customer-facing agents require, and it does so without putting the brand's reputation on the line during the learning period.
On external channels, participate deliberately and instrument everything. Make your catalog discoverable to AI platforms, enable checkout where it preserves your order record, and treat every agent-initiated order as a measurement problem until you know whether it is incremental. Keep identity, attribution, and the post-purchase relationship in-house regardless of which platform originated the sale. Salesforce has removed the excuse that the technology is not ready. What remains is the harder work of governance, incrementality measurement, and deciding how much of your commerce stack to entrust to a single platform in exchange for agents that finally work.
