Key Takeaways
- AI agents are evolving through three phases—researcher, assistant, autonomous buyer—shifting the purchasing subject from humans to machines
- McKinsey projects AI agent-mediated retail transactions could reach $1 trillion in US B2C alone by 2030
- The competitive axis for brands is shifting from "search rankings" to "AI interpretability," making catalogue structuring a core growth strategy
Glu CTO Maps the Evolution of the AI Buyer
Retail shifts as AI agents move from recommending products to completing purchases, forcing brands to make catalogues legible to machine buyers.
ecommercenews.com.auOn April 15, 2026, Sangeeta Mudnal, CTO of AI commerce platform Glu, published an analysis in eCommerceNews Australia examining just how far AI agents are penetrating the purchase process and what this means for retail's structure. After two decades of optimizing for search engine algorithms, brands now face a fundamentally different challenge: becoming intelligible to AI.
Mudnal frames the evolution of the AI buyer in three phases. Phase 1—the "researcher"—is already here. When a consumer asks for "marathon running shoes under $150," AI synthesizes reviews, specifications, and structured data to generate recommendations. Phase 2, the "assistant," is emerging now: agents like OpenAI's Operator handle everything from product discovery to cart creation, with humans only approving the final purchase. Phase 3—the "autonomous buyer"—will see agents restocking essentials and replacing products without any human prompt.
Where the $1 Trillion Market Stands Today
This three-phase evolution is far from theoretical. McKinsey's October 2025 report projects that by 2030, AI agents could mediate up to $1 trillion in US B2C retail transactions. Globally, projections reach $3 to 5 trillion. The report also estimates that ChatGPT alone processes roughly 50 million shopping-related queries daily.
Shopify's own numbers reinforce the trajectory. AI-originated orders grew 11x between January 2025 and March 2026. While AI-referred sessions still account for just about 0.2% of total e-commerce traffic, the annual growth rate stands at 1,079%. The absolute scale remains small, but the growth curve is exponential.
From Search Rankings to Interpretability
The most practically significant part of Mudnal's analysis concerns the paradigm shift in brand visibility.
The era of competing through keywords and backlinks is fading. In an AI-mediated world, the gatekeeper is the reasoning model itself. These models parse structured product data to answer questions, compare options, and validate claims. If a catalogue's descriptions are vague, attributes inconsistent, or data outdated, AI simply cannot validate the product. And what it cannot validate, it will not recommend.
This phenomenon is rapidly being codified as "Generative Engine Optimisation" (GEO). According to Shopify's enterprise blog, pages with complete Product schema are 3.7 times more likely to be cited by AI systems. Where SEO optimized how web pages appeared to humans, GEO optimizes how product data reads to machines.
The Catalogue as Infrastructure: What Shopify's UCP Means
So how is the infrastructure for "machine-readable catalogues" actually being built?
The Universal Commerce Protocol (UCP), announced by Shopify in January 2026 and co-developed with Google, is an open standard that defines a common language for AI agents to connect with merchants—from inventory queries through to payment completion. Over 20 companies including Etsy, Target, Walmart, and Wayfair have already committed to supporting the protocol.
Before UCP, AI agents that wanted to purchase products had to navigate websites, no different from a human shopping in a browser. What UCP enables is direct API-level catalogue access, where agents can execute complete transactions including discount codes, subscription settings, and loyalty credentials. The infrastructure required for Mudnal's "Phase 3 autonomous buyer" is being assembled right now.
Implications for E-commerce Operators
Mudnal outlines four requirements brands should address immediately: structured product attributes that machines can compare, real-time verifiable inventory and pricing, clear relationships between products for contextual understanding, and explicit differentiation that helps models determine relevance.
Her warning that "AI models are being trained on today's data" deserves particular attention. If catalogue structuring lags, the very foundation for how AI understands your brand will be missing. Catching up later is not straightforward—if your products are absent from the training data, you may never enter the recommendation set at all.
As a practical starting point, a GEO audit is recommended: assessing whether your catalogue is interpretable as structured data by AI systems. From there, evaluating Shopify UCP and Google AI Mode compatibility will establish the sales infrastructure for AI channels.
Looking Ahead
"The buyer is no longer just human, and the catalogue is no longer just a website." This single line from Mudnal captures the essence of the agentic commerce era.
What matters most is recognizing that this is not a distant future scenario—the transition from Phase 1 to Phase 2 is already underway. McKinsey's $1 trillion projection, Shopify's 11x order growth, and the rapid adoption of UCP as a standard all point in the same direction. The critical question is whether your catalogue will be part of the conversation when AI moves to Phase 3 and begins completing purchases autonomously.
