business|Jan 15, 2026

When AI can buy: the retail playbook for Agentic Commerce

Alan Descoins

Chief Executive Officer (CEO)

Alan Descoins

Chief Executive Officer (CEO)

Florencia Sanguinetti

Marketing Analyst

Florencia Sanguinetti

Marketing Analyst

For years, AI in commerce has been framed around assistance. Better recommendations, smarter search, faster answers.

But now, a new model is starting to take shape. AI systems are becoming able to act on behalf of the customer, interpreting intent, coordinating multiple steps, and executing transactions across discovery, checkout, and fulfillment. This shift from assistance to delegation is what is being referred to as agentic commerce.

What makes this different is not the novelty of the idea, but the convergence behind it. Standards are emerging, platforms are aligning, and early adopters are moving beyond pilots into production. Google’s introduction of the Universal Commerce Protocol, alongside real-world implementations from companies like Ulta Beauty, Wayfair, Shopify, Etsy, Target, and many more, signals that agent-driven commerce is no longer speculative.

In this post, we’ll break down what agentic commerce actually means in practice, why retail is emerging as its primary testing ground, and what retailers need to have in place to take advantage of this shift.

What agentic commerce actually is (and what it isn’t)

What it looks like for the customer

From the outside, the experience looks simple. A shopper describes what they want to achieve. The system asks a few clarifying questions, then assembles a solution that fits the project, not just individual products.

The interface is deliberately unremarkable. The value is not in the conversation itself, but in what the system is allowed to do next.

What makes it agentic is what happens behind the scenes

What makes this system agentic is not the chat. It is the execution layer behind it.

Behind the scenes, intent is translated into executable steps. Decisions are validated against real inventory, pricing, policy, and fulfillment constraints. Actions are sequenced, checked, and only then allowed to proceed.

This is where many “agentic” demos fall apart. When systems are not connected to operational reality, they can sound helpful while making promises the business cannot keep. In agentic commerce, that gap is not a UX issue. It is a broken commitment.

Drawing the line

In practice, this means being explicit about what qualifies as agentic and what does not.

Once execution authority is introduced, errors stop being cosmetic. They become operational. And success depends as much on alignment across systems as it does on model quality.

This is the line many teams cross without fully realizing the implications.

From responses to actions: what makes agentic commerce possible now

The idea of moving from responses to actions is not new. What makes it viable now is the emergence of shared standards.

Google’s Universal Commerce Protocol is a clear signal of that shift. Rather than acting as another integration, it provides a common layer that allows agents to execute transactions in a controlled and interoperable way.

Practically, this changes what AI systems can safely do:

✔️ complete purchases, not just recommend products

✔️ operate directly within AI-driven environments such as search and agent interfaces

✔️ preserve retailer ownership of identity, customer relationships, and brand experience

This moves agentic commerce out of isolated demos and into shared infrastructure.

At its core, the shift reflects a move away from keyword-driven interactions toward intent-driven ones. As Vidhya Srinivasan, VP and GM of Search Ads and Ads on Google Properties, put it:

Still, standards enable execution. They do not replace the work of integrating data, business rules, and operational reality.

From product recommendation to customer projects

Traditional recommender systems optimize around products. Agentic commerce optimizes around what customers actually want to achieve.

During NRF 2026, this shift surfaced repeatedly across retailer-led sessions, signaling that agentic commerce is moving from experimentation toward execution.

At Home Depot, buying 2x4 lumber is not the goal. Building a deck is. The agent’s role is to understand the project and assemble what is required to complete it. As Jordan Broggi, President of Home Depot Online, explains:

Instead of reacting to what a shopper clicks, agentic systems aim to understand why a purchase is happening in the first place. That shift changes how decisions are made and how value is delivered.

Other retailers apply the same logic in different contexts. Kroger, for example, described shifting from selling ingredients to helping customers arrive at a complete dinner, reducing decision friction rather than just checkout friction.

As this model scales, the challenge for retailers shifts from recognizing intent to supporting execution. As Fiona Tan, CTO of Wayfair, noted during NRF 2026, AI agents won’t just buy products: they will buy projects, such as redecorating a room. Making that possible requires exposing inventory, pricing, delivery constraints, and services in ways agents can reason about autonomously. When those systems remain fragmented, agents stall not because they lack intelligence, but because execution breaks down.

This is why retail has become the proving ground for agentic systems. Intent is often explicit, the stakes are tangible, and failures in execution are immediately visible.

Key insights for retailers

As AI systems move from assisting shoppers to acting on their behalf, retailers are no longer debating if this shift will happen. The real question is what it changes operationally.

The front door may change. The fundamentals do not. In an agent-mediated world, data quality, reliability, and execution become the brand.

Evolution or revolution?

During NRF 2026, agentic commerce was consistently framed not as a sudden disruption, but as the next chapter in retail’s long evolution.

As Josh Friedman, SVP of Ecommerce & Digital at Ulta Beauty, put it, retail has been through similar transitions before: from physical-only to e-commerce, from desktop to mobile, and from channels to omnichannel. Each shift changed how customers entered the experience, while increasing the importance of the systems behind it.

Agentic commerce follows the same pattern. The difference is that the “front door” may no longer be a website, an app, or even a search result. In a zero-click reality, AI agents can research, decide, and transact on behalf of customers without a traditional storefront ever being visited.

For retailers, this is not a loss of relevance, it’s a shift in where relevance is earned.

Agents optimize for outcomes, not brand narratives

Agents don’t care about brand storytelling. They care about price, availability, and service.

This doesn’t make brand irrelevant. It makes it operational. In agent-driven journeys, brand is expressed through questions like:

• Is the product available where and when it’s promised?
• Is pricing consistent and explainable?
• Can fulfillment and service commitments be met reliably?

If an agent promises a Saturday morning delivery and that promise fails, the impact goes beyond a single bad experience. The agent may deprioritize that retailer in future decisions.

In agentic commerce, execution failures don’t just create churn; they may remove you from consideration altogether.

Product catalogs are no longer enough

As agents take on more responsibility, they increasingly operate multimodally. They “see” what shoppers see.

High-quality visuals, structured attributes, and consistent identifiers matter as much as copy — often more. This becomes critical as agents move from recommending products to assembling solutions.

A beauty agent needs to shade-match. A home improvement agent needs to reason about materials and quantities. A furniture agent needs to understand dimensions, finishes, and assembly constraints.

Retailers that still treat images, attributes, and availability data as secondary assets will struggle to participate meaningfully in agent-driven journeys.

Trust is the real scaling constraint

While consumers may trust agents today with low-risk purchases, moving into higher-value transactions requires consistent reliability over time.

Trust here is not emotional. It’s probabilistic. It’s built through predictable outcomes, transparent decisions, and recoverable failures.

For retailers, this makes governance non-negotiable. Human-in-the-loop systems, clear accountability, and exception handling are what allow trust to compound instead of collapse.

Physical stores become the final experience, not the first step

Despite the rise of agent-driven discovery, physical retail is not being displaced. It is being repositioned. Stores are regaining importance as the place for richer, sensory, and high-trust experiences. The moments AI cannot replicate.

In an agentic world, stores become the place where intent is fulfilled with confidence.

These shifts make one thing clear: succeeding in agentic commerce is less about interfaces and more about what sits underneath them.

Things to get right for a great agentic experience

Moving agentic commerce into production is less about innovation and more about readiness. The gap between a compelling demo and a dependable system is filled with unglamorous work: structured data, operational integration, and clear accountability. These are the capabilities that determine whether agents can be trusted to execute at scale.

Below is a short checklist of what you need to focus on:

1. Structured, machine-readable data

Agents reason over data, not copy. In the agentic era, your product data is no longer just a back-office task, it is your brand's primary marketing asset. If your system can't explain why a product meets a specific consumer need in a format an agent can parse, you simply won't make the shortlist.

What to check:

• Shift from keyword-stuffing to Generative Engine Optimization (GEO) by building semantic ontologies like compatibility, certifications, and usage constraints.
• Implement real-time pricing and availability data feeds.
• Enriched metadata must support visual and voice inputs so digital concierges can map photos or project descriptions directly to specific SKUs.
• Maintain unambiguous SKU, GTIN, and MPN mappings across all systems to prevent products from being “hallucinated” out of existence. Verifiable identifiers are key.

2. Frictionless Interoperability

If your system is a silo, agents can't find you. Adopting open standards like the Universal Commerce Protocol allows any agent on any platform to securely interact with your catalog while preserving your brand equity.

What to check:

• Compliance with UCP to enable native checkout inside agent interfaces.
• API-first architecture that supports "machine-to-machine" transactions.
• Secure data-sharing protocols that protect merchant identity.

3. Strategic assortment

AI agents thrive on information density; an "empty" or low-quality digital shelf leads to an invisible brand. Success requires balancing a broad catalog with the high-quality, structured data agents need to justify a recommendation.

What to check:

• The "Extended Mezzanine" model: Use a curated marketplace to fill "white space" so agents can solve complex customer "projects" rather than just finding single SKUs.
• Prioritize "agent-readiness" for high-margin or high-intent items to ensure they aren't filtered out by AI reasoning.
• Maintain strict metadata standards for all third-party items, as agents will ignore products with sparse attributes.
• Ensure extended assortment items follow core brand rules, like in-store returns or loyalty accrual, to maintain trust.

4. Integrated service and loyalty

The agentic experience shouldn't end at the "buy" button. Your agents must have "memory" and access to a unified customer profile to handle returns, apply loyalty rewards, and resolve complex post-purchase queries.

What to check:

• Agents have access to real-time loyalty data (credits, points, tier status).
• Integration with customer service platforms.
• Capability for agents to handle multi-turn interactions across discovery and service.

5. Trust-driven autonomy

Consumers will only delegate "doing" if they trust the outcome. Start with "humans in the loop" and clear guardrails—like spending limits on autonomous purchases—to build long-term confidence.

What to check:

• Secure payment processing via trusted partners.
• Configurable "guardrails" (e.g., spending limits of $50 or $100 for staples).
• Clear transparency on why an agent made a specific recommendation or purchase.

Final thoughts

Agentic commerce is here. The challenge now is execution. Making it work safely, at scale, and in production requires treating accountability as a first-class design constraint from day one.

For retailers, this shift is less about launching new experiences and more about strengthening the foundations that allow agents to act with confidence. In agent-driven ecosystems, operational reliability compounds over time. Retailers that consistently keep their promises are the ones agents will learn to trust and choose.

If you are exploring how to take agentic commerce from pilots into production without breaking trust or operations, now is the moment to start that work. Book a call