In my last column, I discussed the new reality that AI can think and reason at a remarkably high level. Someday we’ll look back and see this as an inflection point in the evolution of technology—a change that will profoundly impact how we think about, organize, and deploy cognitive labor.
When I think about AI’s current abilities, I consider these five key cognitive tasks: perceiving, understanding, reasoning, executing, and creating. Looking at how each is handled in your organization today can help identify opportunities for AI to lighten the load.
Perceiving: Perception is about seeing and making sense of the world around you—and AI can perceive in both the physical and digital worlds.
Self-driving cars are a great example of AI adeptly navigating the physical world. With the ability to accurately identify its surroundings (roads, cars, bikes, pedestrians, signs, traffic lights), AI can make sense of when to stop, go, and change direction.
In the digital world, perception comes into play with computer-using agents, which can perceive and interact with computer interfaces just as humans do. CUAs process raw pixel data to build awareness of what’s happening on the screen and can then interact with that screen—controlling a virtual mouse and keyboard to click, scroll, and type. I expect that businesses will deploy them in fields from sales (assisting in lead generation by automatically filling out forms) to customer service (navigating software applications to find and share information).
Understanding: Understanding goes beyond perception; it’s about seeing patterns and interpreting context. AI’s capacity for understanding means it can interpret, analyze, and generate vast amounts of text data for tasks like translating documents, summarizing reports, and evaluating customer feedback to spot emerging trends. In healthcare, it can lend doctors a hand by interpreting medical images and suggesting possible diagnoses. In finance, it can sift through P&L statements and market data to identify signals that may indicate opportunities or risk.
Case in point: Vodafone deployed an agent that taps into the company’s vast internal knowledge bases to quickly surface product specs, answer legal questions, and more. The telecommunications company’s sales teams use it regularly to respond to RFPs, giving them more time to spend on a task that plays to human cognitive strengths: talking with customers to understand their needs.
Reasoning: I’ve said it before: the ability of AI to reason is one of the biggest technological breakthroughs of our lifetime. Reasoning models solve challenging problems by breaking down a task into parts, analyzing the breadth of the problem and coming up with a plan. Along the way, AI makes lots of smaller decisions, including changing its strategy and reversing course where needed.
Think of a crossword puzzle. You fill in a few words and then find that some of your first answers conflict with the new clues. So you reassess, erase, and try new answers. Reasoning models can now adeptly navigate this iterative process of planning and adapting—and that has big implications for business.
Imagine using that capacity for the multistep research needed to create a competitive analysis, or to produce complex data visualizations that only a data scientist once could. Reasoning AI can perform math at the level of the most skilled humans and has immense potential for scientific discovery. Any knowledge-driven part of a business stands to gain from reasoning AI.
Executing: The fact that AI can execute a task or respond to a prompt on its own is nothing new—it’s the very core of how a prompt-and-response model works. “In-model execution” describes AI’s ability to perform tasks using its internal capabilities. This type of execution is self-contained, meaning the model has everything it needs, including access to necessary data, to complete a task.
But what makes execution so interesting—and something I think will be one of the biggest areas of AI advancement in 2025—is that we’re seeing a second form of AI execution emerge: tool identification and usage. Just like you know to grab a ruler when you need to measure something, AI recognizes when it needs to use external tools to complete a task that goes beyond its inherent capabilities.
Take math, for example. On their own, LLMs are notoriously bad at math. But by enhancing their execution capabilities they’re able to call on outside tools or knowledge sources (like Python capabilities in Microsoft Excel) that enable them to execute complex math formulas. This is unlocking incredible potential for AI to autonomously handle business tasks—from creating images to writing code and visualizing data—that require skills and capabilities beyond their in-model functionality.
Creating: Of all the cognitive tasks, creativity is perhaps the most closely tied to what makes us human. AI is rapidly proving that it can be a powerful creative partner. It excels, for example, at brainstorming: It can tirelessly generate hundreds of product names or taglines (some better than others) for a human team to react to. In business, it can produce everything from concept designs to presentation decks to marketing videos.
In most cases, what AI produces is a starting point; humans elevate it, bringing in their emotional intelligence, nuance, and lived experience. These capabilities are transforming industries in ways that many people, including artists and creators, are understandably still coming to terms with.
Nonetheless, good ideas are good ideas, whatever the source. Not seeking AI’s input and inspiration is now like working with one hand tied behind your back.
What’s next
We’re entering a period of true thought partnership between humans and AI—and still trying to understand where the new division of labor lies. I don’t have all the answers yet, but I do have a few imperatives: As agents begin to handle many cognitive tasks traditionally performed by knowledge workers, organizations will need a new approach to managing them. And as agents and humans collaborate, organizations will need new ways to measure the contributions and performance of each.
Meeting this moment will require a new mindset that goes beyond thinking of AI as a one-to-one human substitute. Early attempts at flying machines were designed to flap their wings, and the first cars were “horseless carriages.” Real innovation comes when we move past imitation.
Despite the incredible advances in AI in just the past six months, many leaders still view it mainly as a means of faster execution. But that’s just the tip of the iceberg. Now is the time to advance AI’s cognitive partnership with humans—and come to terms with its full potential to reinvent how we work.
For more insights on AI and the future of work, subscribe to this newsletter.
