When a consumer asks an AI agent to “find me running shoes under $120 that work for flat feet and arrive by Friday,” something fundamental is changing about commerce. The agent searches inventory, compares options, and checks delivery windows — all without the customer clicking through a single screen.
That shift, from human-driven browsing to AI-driven intent, is the defining commercial transformation of this decade. And according to a new strategy guide published by Publicis Sapient and Salesforce, brands that are not ready to be found, understood, and transacted with by AI agents risk losing revenue to competitors who are.
The window, the report warns, is closing faster than most executives realize.
A $5 Trillion Opportunity — or Someone Else’s
The numbers frame the stakes sharply. McKinsey research cited in the report projects the US B2C retail market alone could see $900 billion to $1 trillion in orchestrated revenue flowing through AI agents by 2030, with global projections for goods reaching $3 to $5 trillion. The report describes this not as incremental growth but as a fundamental reallocation of how commerce moves through the economy.
Consumer behavior is already shifting to match. Half of all consumers now intentionally seek out AI-powered search engines, and 44 percent of those who have tried AI-powered search prefer it over traditional search. ChatGPT alone processes one billion web searches weekly, approximately 350 million of which are shopping-related. Traffic to US retail sites from generative AI browsers and chat services surged 4,700 percent year-over-year in July 2025, according to Adobe data cited by BCG.
Yet the report cautions that current adoption remains concentrated in discovery, not transactions. When AI platforms make product recommendations, 77 percent of consumers still click through to merchant sites to continue their research. Only 23 percent prefer to complete purchases directly within the AI platform. The autonomous checkout era is coming — but it is not fully here yet.
“Agentic commerce is compressing a transformation that would historically have taken a decade into a window of two to three years,” Mohammed AlKhothani, Area Vice President, Salesforce Middle East, said in a statement.
“Businesses across the region that establish agent-addressable capabilities now will compound a durable advantage in data, attribution, and integration that will be exponentially harder to replicate once the market consolidates. That window is open today. It will not remain open indefinitely.”
Jason English, SVP, Publicis Sapient, said, “Agentic Commerce represents a structural shift in how demand will be created, captured and converted.”
“Over the next few years, brands will need to optimise not only for human customers, but increasingly for AI agents acting on their behalf. That requires a new approach to commerce architecture, customer data and experience orchestration.”
The Velocity Problem: This Wave Is Faster
Previous technology transitions offer a useful frame. Internet commerce took roughly 20 years to reach mainstream adoption. Mobile commerce took eight. Agentic commerce, the report argues, will compress that timeline to two or three years.
The reason is structural. AI agents do not require consumers to buy new devices or adopt new behaviors from scratch. They operate across the digital infrastructure — APIs, payment systems, fulfillment networks — that brands have already built. That means the ramp-up time is shorter, the competitive penalties for waiting are steeper, and organizations that hesitate will face what the authors call a “double transformation”: unifying their data first, then making it agent-ready, while early movers have already moved on.
“First movers who establish agent-addressable capabilities today compound a durable data, attribution, and integration advantage that becomes exponentially harder to replicate,” the report states.
Three Kinds of Agents Brands Must Now Serve
The report maps the agentic commerce landscape around three distinct categories of agents, each requiring a different strategic response.
Platform agents — including ChatGPT, Google Gemini, Perplexity, and emerging agentic browsers — function as intermediaries between consumers and merchants. ChatGPT already powers shopping integrations via its Agentic Commerce Protocol and a Shopify partnership. Google is launching “Shop with AI Mode,” which includes price tracking and agentic checkout capabilities. These platforms control discovery, and brands that are not structured for machine readability will not appear in agent consideration sets — effectively becoming invisible before the reasoning loop even begins.
Brand agents operate on merchant-owned assets, serving customers directly within the brand’s ecosystem. Sephora’s Virtual Artist, Zalando’s personalized fashion advisor, Lowe’s LoweBot, and Pandora’s Gemma agent are cited as current production examples. The report positions these as the opportunity for differentiated brand experience, first-party data capture, and direct relationship-building — capabilities that platform agents, by design, do not offer brands.
Personal agents — consumer-controlled AI companions that manage preferences and execute purchases across multiple merchants — represent the most nascent and potentially most disruptive category. They remain in the “autonomous” phase of the report’s maturity model, but the report warns that when personal agents achieve scale, brand choice risks becoming algorithmic rather than emotional.
Why Autonomous Transactions Are Still 2-3 Years Away
Despite the urgency, the report’s authors are precise about what is not yet ready. The Agent Payment Protocol (AP2), which provides cryptographically signed authorization for autonomous transactions, was announced in September 2025 but remains in early-stage pilots, with no broad production deployment. The Universal Commerce Protocol (UCP), launched in January 2026 with backing from Visa, Mastercard, PayPal, Walmart, and Salesforce, establishes an open standard for AI agent commerce — but the surrounding compliance infrastructure is still catching up.
Liability questions remain unsettled: when an agent makes a purchase error, current contract law has no clear answer for who bears responsibility — the consumer, the agent platform, the merchant, or the agent developer. Existing consumer protection regulations, including FTC cooling-off rules and GDPR consent requirements, were written for human-initiated transactions. Fraud detection systems, built around device fingerprints, flag agent-initiated transactions as suspicious.
The report’s recommendation is direct: build for human-in-the-loop transactions now, while preparing architecture for graduated autonomy.
Most AI Programs Are Failing — Here’s Why
Against the opportunity sits a stark failure rate. MIT NANDA research, published in July 2025 and cited in the report, found that despite $30 to $40 billion in enterprise generative AI spending, 95 percent of pilots fail to deliver measurable profit-and-loss impact. Only 5 percent of organizations extract significant returns. Forty-two percent of enterprises abandoned most AI initiatives in the past year, up from 17 percent the year before.
The root causes are not primarily technical. “Workflow misfit” and “organizational resistance” rank above pure technology gaps as failure drivers. Agents deployed without defined escalation paths to human operators, without memory systems tied to retention policies, and without observability infrastructure for reasoning traces are the dominant anti-patterns.
The report introduces a concept it calls the “GenAI Divide” — the widening gap between the small cohort of organizations extracting real value and the large majority spending heavily and delivering little.
The A.C.E. Blueprint: What Agent-Ready Looks Like
The report’s prescriptive framework — called A.C.E., for Agentic Experience, Composable Micro-Apps, and Enterprise Context Orchestration — lays out a three-layer architecture for building agent-ready commerce capabilities.
The first layer focuses on discoverability. Traditional SEO optimizes for human clicks; agent optimization requires what the report calls “model legibility” — ensuring AI can unambiguously parse a brand’s identity, authority, content, and executable functions. Only 36 percent of consumer products companies describe their data as fully structured and machine-readable, and just 33 percent report very consistent product data across channels. Inconsistent data, the report argues, is the new invisibility: agents will route to competitors with cleaner information.
The second layer concerns composable micro-apps: breaking monolithic backends into discrete, atomic API domains — catalog, search, inventory, cart, order — that agents can orchestrate dynamically. The third layer addresses enterprise context orchestration: the governance, memory, access control, and compliance infrastructure that makes autonomous agent interactions safe, auditable, and regulatory-compliant.
Pandora’s implementation, built in partnership with Publicis Sapient on Salesforce’s composable storefront, is offered as a case study: 60 percent autonomous case deflection and a 10-point net promoter score improvement from agent-first service.
The Investment Sequence: Build Now, Prepare for 2027, Monitor Beyond
The report’s recommended investment timeline is structured in three phases. In 2026, brands should build discovery APIs, MCP tool integration, advisory agents, Schema.org optimization, and protocol capability manifests. From 2027 to 2028, the focus shifts to multi-agent orchestration and delegated authorization flows. Fully autonomous transactions, standing intent processing, and cross-merchant basket optimization are categorized as post-2029.
The closing prescription is blunt: organizations have 12 to 24 months to establish agent-optimized capabilities before the market consolidates around dominant patterns. The first step is an audit of current agent-readiness across platform, brand, and personal agent surfaces. The next is deploying a first agent-optimized experience within six to eight weeks.
“Design for humans, optimize for agents,” the report concludes. “The next era belongs to those who master both.”
