By Mohamed Jameel, Managing Director, Votex Insights; author of The Attention Advantage: How to Master Focus, Build Resilience, and Thrive in the Attention Economy.
A document arrives finished. The language is polished, the recommendation looks sensible, and the meeting starts in five minutes. Approving it feels like progress. Asking someone to explain its assumptions feels like holding everyone up.
That small moment deserves more attention than the next announcement about artificial intelligence. It is where a team decides what efficiency means: less unnecessary work, or less thinking altogether.
Research released by Microsoft Research in October 2026 offers a timely reason to reconsider that distinction. In an AI-assisted writing experiment, support that made participants think more carefully improved their work—even though they experienced the AI as less helpful and the task as more demanding. The lesson is not that difficult tools are better. It is that ease can be a misleading measure of value.
For leaders, the opportunity is practical: remove friction that exhausts people, while protecting the few moments of effort that keep understanding, accountability and learning alive.
What the new research actually shows
The October preprint, Metacognitive Scaffolds Create Productive Friction for AI-Assisted Knowledge Work, reports a randomized experiment involving 297 workers and 594 essays. The scaffolds supported goal specification and breaking tasks into parts. The researchers report better writing quality, with a benefit persisting in a short-lag transfer task after the scaffolds were removed, and evidence of better topic recall three days later. Source: https://www.microsoft.com/en-us/research/publication/metacognitive-scaffolds-create-productive-friction-for-ai-assisted-knowledge-work/
Crucially, participants receiving that support rated the AI as less helpful and more cognitively demanding. Feeling supported and becoming more capable were not identical experiences.
This is preliminary evidence, not a verdict on every workplace. The publication is labelled a preprint, the task was writing, and the reported follow-up was short. It does not establish that more friction always improves performance, or that the same approach will improve leadership decisions over months. It does challenge the assumption that the smoothest experience is necessarily the best learning environment.
The timing matters. In a 30 October leadership commentary, Katie McCadden and Andrew Spencer at Harvard Business Impact argued for attention to context, not merely more analysis. A 21 October California Management Review commentary distinguished checking an answer from considering the possibilities it leaves out. These are arguments rather than new controlled experiments, but together they sharpen the management question: what must people still do deliberately when software can finish the document? Sources: https://www.harvardbusiness.org/insight/the-attention-advantage-why-ai-is-making-human-attention-not-intelligence-the-scarcest-resource-in-leadership-part-one/ and https://cmr.berkeley.edu/2026/10/blind-overseers-ai-and-the-erosion-of-judgment/
Remove bureaucracy, retain understanding
Not all friction deserves protection. Re-entering the same figures, searching for a missing file, waiting for an unnecessary signature: these consume attention without improving the decision. Automate them where appropriate.
But explaining why a recommendation fits this situation is different. So is checking the source behind a consequential number, or listening to a colleague who sees a risk the summary missed. These are not administrative obstacles. They are part of the work.
The distinction is particularly useful for small teams, where the same person may prepare, review and execute a decision. An elegant output can hide the absence of an independent check. Adding another approval box does little if the reviewer lacks either the time or the knowledge to challenge it.
Consider a hypothetical Maldivian hospitality team using AI to prepare a guest-service improvement plan. The document recommends faster response targets. Before approving it, the manager asks whether staff can meet those targets during transfers, whether language needs were considered, and whether speed is the right measure for a sensitive complaint. Nothing in this example requires rejecting AI. It requires restoring context to the decision.
Start before the first prompt. Ask the person commissioning the work to write three sentences: what decision this will support, whose interests must be considered, and what would make the answer unacceptable. That brief makes the purpose visible before polished language can define the problem for everyone.
At review, ask for a short explanation in the author's own words: what supports the recommendation, what remains uncertain, and what credible alternative was considered. This is not a memory examination or a demand to recreate everything unaided. Source documents should remain available. The aim is to discover whether the person can connect evidence to action.
Then match checking to consequences. A routine internal reminder and a proposal affecting employment, money or safety should not face the same review burden. For consequential decisions, name a qualified reviewer, verify material claims against original sources, and record unresolved uncertainties. Where relevant, obtain appropriate professional advice. Accountability should have a name, not merely a button.
Try a small test, not another policy manual
Choose one recurring, low-risk workflow for a two-week trial. Keep the existing quality checks, but add the three-sentence brief and the short explanation at review. Tell the team this is an experiment in better work, not surveillance or a test of who deserves to use AI.
Track more than completion time. Note the corrections needed after review, the assumptions caught before action, and whether another colleague can understand why the recommendation was chosen. Ask participants which steps helped and which merely added work. Keep sensitive material out of unapproved tools.
This protocol is a practical proposal, not a package validated by the Microsoft experiment. Its value must be tested in your setting. If a step produces no useful insight, simplify it. If review repeatedly uncovers the same knowledge gap, provide training or change the workflow rather than blaming people for failing a process they were never equipped to perform.
Leaders also need to change what they praise. If instant answers earn recognition while thoughtful questions earn impatience, a written commitment to responsible AI will achieve little. Make room for someone to say, “I need to verify that,” without treating the pause as incompetence.
Mindful leadership belongs here in a modest, concrete way: notice the urge to approve because the document looks finished, then return attention to what the decision requires. A pause is not a substitute for expertise, adequate staffing or good governance. It can help us remember to use them.
The aim is not to make work harder. It is to stop confusing the removal of effort with the creation of value. Automate the repetitive burden. Preserve the human work of setting direction, examining consequences and learning from what happens next.
That is the attention advantage worth building: not simply finishing more, but knowing what deserves to be finished, checked or reconsidered. For a deeper exploration of focus and resilience, see The Attention Advantage: https://www.amazon.com/dp/B0HP6FLWYL. To explore practical attention and leadership development for your team, visit Votex Insights: https://www.votexinsights.com.
