What every digital executive needs to know to navigate the evolving AI landscape.
AI used to mean clunky chatbots and fuzzy recommendations. Then ChatGPT cracked things open. Almost overnight, AI went from background noise to boardroom priority, spawning a rush of tools, pilots, and promises.
Now, every vendor is "AI-powered." One pitches a personalization engine, another touts agent-based automation. Internal teams spin up copilots or integrate with Gemini. But beneath the flurry of demos and decks, the fundamentals remain confusing: What’s a model versus a tool? Is embedded AI really doing anything? And are autonomous agents the next frontier or just the next distraction?
For digital leaders, these distinctions are more than semantics. They shape budgets, architectures, and roadmaps. And in a fragmented, fast-moving ecosystem, clarity is more than helpful— it’s a competitive advantage.
Most AI capabilities in commerce today fit into one of four layers: foundation models, tools, embedded features, and agents. These layers aren’t strictly hierarchical—they overlap and interact, each playing a different role in the stack. But knowing which layer you’re working with helps define its risk, potential, and operational impact.
Many of the worst AI decisions come from treating these layers interchangeably. The solution is clear categorization—and brutally honest evaluation.
Foundation models like GPT-4, Gemini 1, and Claude are the intellectual engines of modern AI. They don’t come with UIs or templates; they require wrapping, tuning, and often significant engineering effort to apply usefully. But they offer raw power that can transform how a business operates.
For example, a brand with millions of SKUs might use a model to automatically generate product descriptions—tailored by category, tone, and channel—at scale. But without the right prompt structure, governance process, or validation step, this “solution” becomes a liability.
That’s why understanding what model underpins a tool or platform matters. A vendor building on GPT-4 may be better at language, but worse at reasoning than one using Claude. A closed model might be easier to deploy, but harder to customize. Leaders don’t need to know the math—but they do need to understand the tradeoffs.
As you evaluate the vendors and solutions to include in your stack, there are a few questions you should ask. Things like. “What foundation model do you use?” and “Why?” If they don’t have an answer, they don’t understand their own product— and that’s a red flag.
Tools are where most people experience AI firsthand. Interfaces like ChatGPT, Gemini, Writer, or Jasper wrap foundation models in usable workflows that solve specific problems, whether it’s writing product descriptions, optimizing SEO, or assisting customer service. But tools vary wildly in polish, flexibility, and alignment with real-world commerce needs.
Many execs fall into one of two traps: either treating these tools as magic wands, or dismissing them as toys. The truth lies in between.
A good tool can multiply the productivity of marketers, merchandisers, and support teams. For example, a global brand using Writer can maintain tone and terminology consistency across dozens of markets while cutting copy production time in half. But this only works if the tool understands brand voice, has access to the right product data, and is integrated into daily workflows.
When weighing the options for AI tooling, don’t pilot them in a vacuum. To really determine their worth, you need to test them in the messy, real context of your business, using your data, your teams, and your constraints.
Some of the most impactful AI in commerce isn’t flashy—it’s invisible. Embedded AI isn’t new, showing up in the platforms brands already use to silently power personalization engines, search optimization, pricing algorithms, and even product taxonomies. When vendors like Algolia, Contentful, and Salesforce Commerce Cloud weave machine learning into their core features, your business can get real results without the need for in-house AI talent or prompt engineering expertise.
The danger is assuming all embedded AI is valuable. Not every “AI-powered” claim holds water. Some vendors simply rebrand existing rules-based logic. Others deploy serious ML models but fail to surface their impact clearly. Digital leaders must learn to interrogate these systems.
For example, a product discovery platform might claim it uses AI to rank search results, but how do you know? It’s imperative to evaluate embedded AI not on what it promises, but on what it actually learns from and how it adapts. Intelligence without adaptation is just automation.
Before you put too much faith in embedded tools, ask vendors tough questions: What’s actually powering this feature? Is it learning, or just rules-based logic? How is its effectiveness measured? Is it learning from click data? From inventory levels? From customer segments? The answers can cut through the buzzwords and reveal what’s truly valuable, because if it isn’t adapting in real time, it’s not intelligent.
The final AI layer is also the one with the most hype— and the most misunderstanding. As people grow more comfortable with AI tools, agents are becoming the cool new kid on the block. While tools help humans work faster, and embedded features make platforms smarter, agents are designed to act autonomously. These systems don’t just respond; they decide.
Unfortunately, while the term implies autonomy, many so-called “agents” are just chained API calls or scripted workflows with a language model stuck in the middle.
But real agents are emerging. In commerce, they could autonomously create product bundles based on real-time inventory, customer behavior, and competitor pricing, then launch campaigns across multiple channels without human involvement.
Today, agents are mostly early-stage (and some are little more than complex macros wrapped in AI branding). Others show real promise but face limits in memory, context, or security. Still, the direction is clear. Agents are coming, and while they may not replace humans, they will increasingly own repeatable tasks across marketing, merchandising, and customer service. Smart leaders are exploring them now, with guardrails in place, to be ready when they mature. Because waiting too long means missing early learnings that could shape future capabilities.
Even though we’re in the early phases of agentic experiences, there are steps you can take today to be prepared. Don’t aim for fully autonomous agents right now, instead, start by designing “copilot loops” — an AI agent proposes the next best action, a human reviews and approves, and the system learns from the outcome. As trust and capability expands, more of the experience can be handed to an agent.
Strategic AI adoption isn’t about chasing innovation—it’s about orchestrating clarity. And amid the noise, digital executives need a simple but powerful decision framework.
First, identify the layer you’re evaluating: is it a model, a tool, a feature, or an agent? Then you can move on to the questions that determine its worth to your business: What outcome(s) does it drive or problem(s) does it solve? Where and how does it integrate with existing systems and architecture? Is it secure and governable? How do you measure its impact?
Not every AI investment needs to be a moonshot. Often, the biggest gains come from using mature, embedded AI to optimize what already works while reserving more experimental energy for targeted pilots with tools or agents. AI isn’t one product or strategy—it’s a connective tissue, connecting systems, content, and decisions.
Ultimately, competitive advantage won’t come from being the earliest adopter. It will come from being the most intentional—knowing where to automate, where to augment, and where to differentiate. The leaders who understand how AI works—not just what it does—will be the ones who define how commerce evolves.
Learn more about where AI will impact commerce experiences in our latest Masterclass, AI for Commerce: Foundations, Aspirations, and Hype.
Leigh Bryant
Editorial Director, Composable.com
Leigh Bryant is a seasoned content and brand strategist with over a decade of experience in digital storytelling. Starting in retail before shifting to the technology space, she has spent the past ten years crafting compelling narratives as a writer, editor, and strategist.