The artificial intelligence conversation in retail has reached a critical inflection point. After years of breathless predictions about artificial general intelligence being “just around the corner,” the industry is waking up to a more nuanced reality. And, David Dorf, Head of Retail Industry Solutions at AWS, isn’t pulling any punches when it comes to where we’ve been and where we’re actually heading.
In a recent conversation on Omni Talk’s 5 Insightful Minutes, Dorf laid out a frank assessment of the AI landscape that every retail technology leader needs to hear. His central thesis? We got ahead of ourselves with large language models, and the path forward requires a fundamental shift in how we think about AI deployment.
The LLM Plateau: Why Bigger Isn’t Always Better
Just a couple of years ago, the prevailing wisdom suggested that simply scaling up data would lead to exponentially better AI results. That assumption is now being challenged by the reality of diminishing returns. “LLMs are tremendously powerful, but the idea that we can continue to just scale up the data doesn’t actually lead to better results,” Dorf explains.
The numbers back this up. LLMs will continue to improve incrementally, but without a fundamental architectural breakthrough similar to the Google transformer that launched the current generation of language models, exponential gains remain elusive. This isn’t a failure of the technology, either. It’s a maturation of our understanding about what these tools can and cannot do.
Domain-Specific Models: The Next Frontier
Rather than chasing the dream of artificial general intelligence, Dorf advocates for a more pragmatic approach: domain-specific models designed for particular use cases. These specialized models have already proven themselves in finance and healthcare, and retail is the next logical frontier.
AWS’s recent announcement of Nova Forge represents a significant step in this direction. The product takes a unique approach by providing customers with a half-trained frontier model that they can finish training with their own specific data. For retailers, this opens up powerful possibilities. For example, imagine a tightly focused LLM that deeply understands your product catalog, knows your brand voice, and can deliver results that generic models simply cannot match.
The cost implications are equally important. Domain-specific models offer a path to lower operational costs while delivering better, more relevant results. In an industry where margins matter, this combination of improved performance and reduced expense could be hard to ignore.
Agentic Commerce: Progress With a Reality Check
If you felt overwhelmed by the flood of agentic commerce announcements at the end of 2025, you weren’t alone. Dorf describes it as more of a “tsunami” than a flood, with major retailers like Walmart and Target rushing to implement solutions during the crushing month of November.
His assessment, like our very own Chris Walton’s a few weeks ago, is measured: “Fantastic progress, but let’s put the cake back in the oven for a little longer.” Some implementations felt clumsy, and the technology hasn’t yet reached the seamless experience that will drive widespread adoption. However, the direction is clear, and early indicators are promising. Amazon’s Rufus, for example, is showing an uplift in sales from shoppers who use the tool compared to those who don’t.
Because key questions still remains, questions like: Did the answer engine get the customer to buy when they wouldn’t have otherwise? Today, that answer is largely no, primarily because of disjointed experiences. But if retailers can get the experience right, agentic commerce has the potential to deliver genuine incremental sales.
Answer Engines: The New Search Battleground
As bots now officially outnumber humans on the internet (a threshold crossed just last month) retailers face a new imperative. Just as product listings in search results became non-negotiable, appearing in answer engines is becoming equally critical.
But this shift brings familiar tensions. Running answer engines is expensive, and consumer subscriptions alone won’t cover the costs. Advertising is coming, starting with sponsored prompts that Amazon and Walmart are already testing. This introduces the same trust challenges that have long plagued traditional search engines.
Answer engine optimization companies are also emerging, figuring out how to manipulate LLMs just as SEO specialists learned to game search algorithms. Retailers have already discovered that ChatGPT favors top-10 lists, prompting a wave of blog content designed to game the system. It’s an echo of battles that date back decades, a reminder that the more things change, the more some dynamics remain constant.
Shopping Agents: Starting With Delivery
The most intriguing near-term development may be third-party shopping agents that automate specific tasks through subscription services. Dorf predicts that delivery companies like Instacart and DoorDash will lead this charge. They already aggregate products from multiple sources and have the delivery infrastructure in place. The natural evolution is to offer personalized agents that understand your preferences, favorite restaurants, typical grocery items, and shopping patterns.
The real value comes from personalization and outcome-based shopping. Instead of searching for individual products, imagine telling an agent to “plan a birthday party with a pirate theme” or “arrange a dinner party for eight with a French bistro theme,” and having it handle everything from sourcing to delivery. This shift from product-focused to outcome-focused shopping represents a fundamental change in how consumers might interact with retail.
Interestingly enough, UBS is also predicting that grocery will be the first industry significantly impacted by agentic commerce, which is the exact opposite of how traditional e-commerce evolved. The high frequency, low consideration nature of grocery shopping makes it an ideal testing ground for agent-based purchasing.
Internal Operations: Where the Crystal Ball Gets Clearer
While consumer behavior remains unpredictable, internal retail operations present a clearer opportunity for AI agents. Companies are laser-focused on efficiency, making this the sweet spot for AI deployment. Retailers will increasingly deploy reasoning-capable agents to assist with merchandising, marketing, and supply chain processes.
AWS announced three new frontier agents at its recent re:Invent conference covering developers, security, and cloud operations. These horizontal agents that work across industries will eventually give way to industry-specific agents tailored for retail functions. It’s not hard to imagine SaaS subscriptions eventually including agents that help users run the applications more effectively.
The partnership between AWS and Visa around agentic payments points to particularly compelling B2B applications. The entire purchase order to payment cycle, from inventory receiving, invoice matching, policy verification, to payment execution, can all be automated by agents, eliminating manual checks and streamlining processes that have remained stubbornly inefficient.
The Path Forward
Net/net, David Dorf’s predictions for 2026 aren’t about revolutionary breakthroughs or paradigm-shifting technologies appearing overnight. Instead, they reflect a mature understanding of where AI can deliver genuine value right now. Domain-specific models will replace the pursuit of artificial general intelligence. Answer engines will continue evolving while grappling with familiar advertising and trust tensions. Shopping agents will start with practical applications in delivery services before expanding to broader use cases. And internal retail operations will see the most immediate and measurable impact.
For retail technology leaders, the message is clear: focus on pragmatic implementations that lower costs and solve specific problems rather than chasing the next big headline. The future of AI in retail won’t be built on hype. It will be built on domain expertise, personalization, and a relentless focus on outcomes that matter.
And, if you enjoyed this post and our talk with David, the conversation will continue at NRF 2026, where AWS will be in booth 4438, and at the Retail ROI Super Saturday event, where industry leaders gather to discuss not just technology but also giving back to communities in need. Because ultimately, the best technology serves not just business objectives but broader human purposes.
Watch the full 5 Insightful Minutes episode above to hear David Dorf’s complete insights on how he thinks the world of retail AI will unfold in 2026.
Omni Talk® is the retail blog for retailers, written by retailers. Chris Walton and Anne Mezzenga founded Omni Talk® in 2017 and have quickly turned it into one of the fastest growing blogs in retail.