Rethinking AI in Knowledge Work: From Assistant to Tool for Thought

By Advait Sarkar, Senior Researcher; Sean Rintel, Senior Principal Research Manager; Leon Reicherts, Researcher; Lev Tankelevitch, Senior Researcher; Pratik Ghosh, Senior Research Designer; Richard Banks, Principal Design Manager; Payod Panda, Design Engineering Researcher; Martin Grayson, Principal Research Software Development Engineer

In today’s workplace, knowledge workers face a paradox. We are surrounded by powerful AI systems that promise to make us more productive by summarizing our emails, drafting reports, analyzing data, even building presentations. Yet instead of feeling more capable, many find themselves wondering: am I still doing the thinking, or just validating what a machine has produced?

At Microsoft Research, we are investigating a different path: designing AI not to replace thought, but to deepen it.

The risk of outsourced thinking

Consider a typical day in the life of a 21st-century knowledge worker. You arrive at the office and face an inbox overflowing with messages. AI offers to summarize and draft responses. A blank report awaits, so you generate a first draft with AI. Next comes data analysis that AI again completes. Finally, you need slides, so AI creates a presentation in minutes. Efficient, yes. But when you step back, you worry that you might become, as Advait Sarkar references in his upcoming talk at TEDAI Vienna, a “professional validator of robots’ opinions.”

This mode of work, where a knowledge worker no longer engages with the materials of their craft, is already here. And it could have consequences. Research in the field has shown mixed outcomes: some studies suggest that, when used in the right way, AI can actually enhance creativity and critical thinking, while others show the opposite. Our own findings highlight the risks. Studies show that using generative AI can lead knowledge workers to produce a narrower range of ideas, put less effort into critical thinking, and retain less of what they write or read. Our survey-based study suggests that when people view a task as low-stakes, they may not review outputs as critically; however, when the stakes are higher, they naturally engage in more thorough evaluation. Even memory – the ability to recall what we’ve worked on – can be affected when the processes of work are intermediated by AI. In other words, we may get the job done faster, but at the risk of weakening our cognitive “muscles.”

Reimagining AI as a partner in reasoning

At Microsoft Research, we are exploring how AI can be used to keep humans cognitively fit – designing tools that push, provoke, and expand our reasoning. The work we presented at CHI 2025 builds on this vision.

Our team developed prototypes that integrate AI into everyday workflows in ways that support human cognition rather than displace it.

For example, in addition to assisting with drafting the report, AI might highlight tensions between sources, surface alternative interpretations of data, or ask probing questions that spark deeper exploration. The aim is not just to produce an output, but to help the human worker reach new insights.

Protecting cognitive skills that make us human

The distinction is subtle but profound. AI as an assistant is about speed and efficiency: finishing tasks more quickly. AI as a tool for thought is about depth: creating outcomes of higher quality, asking more pertinent questions, and exploring unknowns. This matters because knowledge work isn’t just about execution; it’s about judgment, sense-making, and creativity. If AI intermediates these processes, we risk becoming what Sarkar refers to in his TED talk as “intellectual tourists” – visiting ideas but never truly inhabiting them.

Instead, tools for thought can preserve and even strengthen the habits and capacities of mind that make complex work possible. Tools for thought encourage us to wrestle with ambiguity, reflect on our reasoning, and stay intellectually engaged. Just as a microscope extends our vision, or learning algebra allows us to conceive of and tackle mathematical problems in new ways, these systems can extend our thinking.

Choosing a future that elevates thought

The question is not whether AI will shape the future of knowledge work – it already has. The real question is how we choose to shape it in return. If we allow AI to think for us, we risk hollowing out the very skills that make progress possible. But if we design AI to think with us, we can cultivate a generation of workers who are more creative, more reflective, and more capable than ever before.

The choice extends well beyond business – fields like education, science, and policy making all stand to benefit from AI that sharpens rather than dulls our intellect. In time, these systems will be as integral as spreadsheets or search engines, but only if we design them with explicit attention to how humans think, not just the artefacts that their work demands they produce.

At Microsoft Research, we see this as a defining opportunity of our time: to ensure that AI does not only accelerate work but also elevates human thought. The choices we make today will determine whether the next era of knowledge work is one of intellectual complacency or of unprecedented insight.