In December 2025, the profession of software engineering profession changed profoundly. State-of-the-art agentic engineering tools developed by hundreds of companies in the preceding 18-24 months were suddenly paired with a new model, Anthrophic’s Claude Opus 4.5.
While the score improvements to the usual AI benchmarks seemed modest, the gains in utility were immense. We crossed an invisible, but very real, capability threshold that none of the benchmarks could predict, and it changed the fundamentals of software engineering forever.
AI systems gained a new level of trust, based on true merit. What followed was a rapid shift from a mindset of “AI is a tool that helps me write code that I must review carefully” to “I can orchestrate teams of agents that fully automate the majority of the software development process.”
Change propels innovation
This shift happened because the agentic infrastructure (prompts, tools, data) for the software development domain matured enough to unlock the new model’s capabilities. Although, “mature” may be a misnomer in this context. As the productivity of software developers and agents skyrockets, we see that many processes and technologies need to be rethought from the ground up. This has led to a dramatic increase in the pace of innovation in agentic infrastructure for software development, whose gains will compound as smarter, faster and cheaper models emerge. In short, things will change weekly, and begin to mature once again.
Many industries, such as retail, have been unaffected by AI. Sure, CPGs and retailers might roll out horizontal chatbots like Claude, Copilot, or ChatGPT, but those initiatives will not have a profound impact.
Raw intelligence is not what is holding industries like retail back. Instead, it’s the lack of specialized agent infrastructure for vertical AI. AI agents cannot run a business on slide decks and scattered text documents in shared folders. While the AI models are smart enough, the retail industry needs to build a solid foundational infrastructure to unlock AI’s full potential.
Over the past decade, the retail industry has been riddled with digital transformation projects with noble goals, but often questionable outcomes. The organizations that have been successful, however, adopted many of the most important components of the vertical AI stack.
When AI is built vertically, on a foundation of data that is contextualized through industry-specific semantic modeling, actions not only arrive faster, but with accuracy that can be validated. This allows every output to connect back to the underlying data, business context, and logic behind it. This is a vertical approach to retail AI, where enterprises can establish a continuous learning loop that stays responsive to real-time signals across stores, channels, regions, demographics, and more. When this loop gets “good enough” to be trusted to run fully autonomously is when we will see retail benefit from the same, profound opportunities we have seen in the software engineering industry.
A robust data foundation is key
Retail brands face pressure to monitor performance at the most granular levels, down to the SKU and store level across an entire portfolio. In practice, it’s challenging to address anything beyond the most pressing issues at the top retail accounts due to data overload. But with vertical AI, CPG manufacturers can monitor granular real-time signals across individual stores and products, as well as strategic groupings, to optimize inventory flow, refine promotional strategies, curate store assortments, expand margins, and more.
Structured retail data lays the foundation for vertical AI for retail, where streamlined, connected, and enriched data ensures crucial accuracy and consistency when recommending actions tied to real ROI.
Dave Nolen, VP of Category Leadership and Shopper Insights at Kraft Heinz, commented that “AI can speed up retail collaboration but only if the right foundation is in place. Companies like Crisp help us achieve hyper-localization at scale, by securing the clean and structured data necessary to act.”
Transparent logic chains are essential for trust
Speed without accuracy introduces significant risk, underscoring the pressing need for logic chains in enterprise AI. Especially in commerce, the wrong decisions can quickly affect revenue, inventory, and critical partnerships. Agentic AI must be built on clear logic chains, not just outputs. Importantly, every recommendation should be traceable back to its data inputs, algorithms, and applied business rules. And, outputs should reflect retailer-specific constraints product portfolio hierarchies, and operational realities, rather than general pattern recognition.
Without logic chains, AI remains observational, requiring teams to spend an indeterminate amount of labor validating outputs and connecting them to action.
Context compounds over time
AI without memory or an understanding of user preferences is limited to one-off analysis and is difficult to scale. From day one, vertical AI prioritizes learning your industry and company language, terminology, KPIs and how success is measured. Retail, category, and channel nuances are captured, in concert with how teams make and act on decisions.
Over time, collective memories become building blocks that lead to insights faster, in a way teams can share and collaborate around effectively. This is what separates vertical AI (AI that understands retail at its core) from general-purpose tools that reset with every session.
Franz Oliveira, Global Director of Analytics at ZURU, shared that “Crisp’s AI was very custom-made to my need to answer a retail-specific problem. Other AI just gives me generic analyses and recommendations.”
Building trusted, productive partnerships with vertical AI
Kraft Heinz’s Away From Home team is leveraging vertical AI within its foodservice distribution channel to accelerate decision-making and execution. Jeff Garde, Distributor Development at Kraft Heinz Away From Home, shared that he “was able to develop a strategy with my top distribution partner in 10 minutes with an AI platform.”
Schwan’s Company is leveraging vertical AI to operationalize retail intelligence, helping teams protect revenue, uncover risk sooner, and act with greater speed and precision. Ben Martel, Category Manager at Schwan’s, described the impact: “AI Agents have streamlined my reporting process, saving significant time each week, while delivering advanced insights through interactive visualizations.”
Martel continues, “AI has enabled me to quickly identify high-performing items with low distribution as potential growth levers, as well as spot low-performing items that are at risk” – helping his team make precise, data-driven assortment recommendations tied to real store outcomes.
“Doing more with more”
So, what does retail optimization at scale look like? The current phase of retail AI is making teams comfortable with its capabilities. After working through the “first day” feeling – when the contextualized knowledge and personalized flow set in – comes the testing phase, followed by implementation at scale.
What comes next is that AI will begin to surface too many problems and opportunities for teams to act on across stores and SKUs, requiring manual intervention. This is where agent-to-agent communication will build a bridge, with tangible ROI gains and many learnings to follow if implemented effectively.
The retail industry is defined by fierce competition, and that competition will only continue to evolve. Extreme responsiveness across the supply chain is essential. The question is not whether companies will become fully resource-optimized, but whether they will be equipped to deliver in an ecosystem where major players continue to bid dynamically for consumers’ loyalty.
The future of retail can become a prosperous greenhouse of innovation and consumer delight with the right vertical AI foundation in place.
