Your AI tools should be working together better

fastcompany.com

No one buys a dishwasher on Tuesday, then wakes up Wednesday thinking: “ You know what I could really use? Another dishwasher.” But that’s essentially how many e-commerce systems still operate. A shopper may spend weeks researching dishwashers—visiting a retailer’s website, using the mobile app, and eventually going to their local store to narrow down the options in-person—before making the big purchase. Yet the next day, they return to the website, or open an email or text from the retailer, only to find themselves encouraged to buy…another dishwasher. It’s archaic, a little ridiculous, and often the result of a surprisingly simple problem: The systems that registered the shopper’s interest aren’t working from the same context as the one that registered the purchase. What would be more useful is to recognize the customer’s needs have changed. Maybe now they need installation guidance, maintenance tips, detergent tabs, or something else complementary to what they just bought. But they almost certainly don’t need another dishwasher. AI’S NEXT CYCLE The first wave of GenAI showed what models can do with massive amounts of information, from answering natural-language questions (“Which dishwasher features matter most for a family of four?”) to summarizing complex materials to creating content. But generating polished responses is different from understanding the big picture—including individuals’ actions, context, and the relationships between signals —and using that to determine what happens next. To bridge that gap, there’s rising interest across industries in the concept of reasoning engines. At a basic level, reasoning engines are AI systems that perform logical inference based on real-world inputs, not just questions or prompts people give them. And rather than considering one interaction or data point in isolation, they can reason across multiple signals and channels. This way, recommendations, including at the individual level, are most likely to land. STRONG LIMBS, BUT NO BRAIN Sprawling tech stacks have made the need for overarching reasoning more pronounced. This didn’t happen overnight. Organizations have spent years assembling and deploying specialized technologies, each designed to do one thing very well. For example, in e-commerce, search interprets what a consumer is actively asking for (“a quiet dishwasher”), while recommendations predict and display products that might interest a given shopper. AI agents and chatbots interact with shoppers conversationally, and marketing systems manage communications across channels like email and SMS. In-store POS systems know what happened offline, including that the customer purchased a dishwasher. And so on. AI is making these individual systems more capable. But greater intelligence at the point-solution level doesn’t automatically create a more intelligent experience overall. Think of specialized systems as highly capable limbs. Strengthening each one doesn’t solve the coordination problem if there’s no central nervous system. In other words, if your left leg doesn’t know what your right is doing, turbo-charging them may just make you stumble faster. That’s why a reasoning layer matters. It acts as the brain or connective intelligence, letting systems share context and coordinate decisions, so the whole body moves together. REAL-WORLD REASONING Across sectors, companies and their customers are starting to see the benefits of connected intelligence. In banking, for instance, agents reason across risk signals , gauging what’s significant and what’s not, while reducing manual work across traditionally fragmented know-your-customer processes. And in supply chains, companies are using central, connected intelligence to link planning, procurement, and logistics decisions, moving from “ firefighting to orchestration .” Retail offers another useful illustration, where innovative brands connect breadcrumbs across the shopper journey—from website and in-app behavior to loyalty program status—to inform what someone sees next. An online shopper checking out men’s Hawaiian shirts, for example, might ask the site’s AI assistant later that day for vacation outfits. Reasoning across the shopper’s browsing behavior and purchase history, the agent highlights men’s tropical gear in that person’s preferred size and colors. DON’T START WITH THE TECH These examples show some benefits early adopters are achieving. Others can, and should, begin laying the foundation. Think about it. Companies built websites before e-commerce became ubiquitous. Similarly, they shouldn’t necessarily wait for reasoning technology to reach widespread adoption before building the architecture to support it. Here are ways to get ready: Break down context silos: Identify where important signals live and whether relevant systems can access and share them. Whip underlying data into shape: A reasoning layer isn’t magically useful if the information it’s reasoning over is incomplete, inconsistent, or inaccessible. Architect for orchestration: When adding AI agents or other applications, consider how they’ll connect with existing systems rather than evaluating them just on standalone capabilities. Design around the user’s journey: Customers don’t experience channels and point solutions; they experience one brand. Technology architectures should reflect that. TOMORROW’S MAINSTREAM CONCEPT Reasoning tech isn’t new. It’s been part of the AI and data science lexicon for several years now. A 2022 Google Research paper introduced chain-of-thought prompting and voices like Yann LeCun, former chief AI scientist at Meta pursued a different architecture for how machines can reason and plan. Though largely still a technical concept, there’s precedent for how quickly that can change. Flash back to 2017, for instance, when a Google Research paper introduced a new kind of neural network, the transformer. The concept was obscure then, but the paper has since been cited extensively. The architecture it described became the foundation nearly every modern LLM is built on—the “T” in ChatGPT. Now, as AI makes individual solutions more capable, and the need for orchestration more important, reasoning engines may be on a similar trajectory. And as we think back to that dishwasher, it’s clear why. That shopper doesn’t know or care which system powered which interaction. They just expect the brand to understand the context they’ve already provided. (Rinse and repeat!) When enterprise tech behaves like that, we will have reached the next era of AI, where individual systems aren’t just smarter; they work together. Then, technology itself becomes less visible, and things simply work the way consumers expect they should. Kevin Laymoun is chief customer officer and chief revenue officer at Constructor.