> ## Content Index
> Fetch the complete content index at: https://www.mikebakerhq.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# AI Is Making Us Better at Working Alone. I’m Not Sure We Understand What That Changes.
- URL: https://www.mikebakerhq.com/writing/ai-working-alone/
- Published: 2026-08-26T03:33:14.000Z
- Updated: 2026-08-26T03:33:14.000Z
- Description: AI is making individuals more capable — and quietly changing how much we rely on each other at work. What leaders should watch as the tools spread.
- Author: Mike Baker
- Tags: Human First

I use AI every day, enough that I have started noticing how it changes the way I move through work. A few years ago, if I got stuck on something, I usually had to involve another person. I might walk into someone’s office, send a quick email, call somebody who knew more than I did, or hand over a rough draft and ask what they thought. Now I often start with AI. I ask it to help me sort through an idea, find the weak spots in an argument, organize information, or help me understand something before I take it to another person.

Most of the time, that is useful. I get moving faster, and someone else avoids an interruption they did not need. But I have started wondering what else changes when those small interactions disappear.

That question stayed with me after I read a new workplace study this week. Researchers looked at Microsoft 365 activity across 11 large international companies after employees gained access to Copilot. More than 40,000 people were included in the dataset, and 7,831 of them became frequent users, which the researchers defined as using Copilot more than 100 times during the first 20 weeks.

Among those frequent users, activity in productivity applications like Word, Excel and PowerPoint increased 21.2%. Communication activity increased too, but by 7.1%. The researchers also found a decline in some small-group email activity, with slightly fewer people involved in those exchanges and fewer rounds of back-and-forth conversation.

Those are not dramatic numbers, and the study is still a preprint. It has not yet been peer reviewed. The researchers also acknowledge that activity inside Microsoft 365 is an imperfect way to measure something as complicated as productivity or collaboration. I would not build policy around any single percentage in the paper.

What caught my attention was the direction of the change. People who used AI heavily appeared to become more productive while relying a little less on other people for parts of the work.

That feels believable because I see it in myself. I can do more alone than I could a couple of years ago. AI helps me understand an unfamiliar topic before I ask an expert for help. It helps me work through a document without asking someone else to summarize it. It gives me a first reaction to an idea before I bring it to another person. A lot of workplace friction deserves to disappear, and I am happy to see it go. Nobody should spend twenty minutes digging through old email for a file that software could find in seconds, and I have sat through enough meetings that existed mainly because information had to travel from one person to another.

But work has always contained another kind of information that is much harder to measure. Every organization has things people know because they have been there long enough to know them. Someone remembers why a process exists even though the original decision was never documented well. Somebody else remembers what happened the last time the organization tried something similar. A person working close to patients or customers notices a problem long before it appears in a report. A newer employee asks a basic question and accidentally exposes something everyone else stopped questioning years ago.

A lot of that information moves through ordinary conversations. You walk into somebody’s office to ask about one thing and learn about something else. You call someone for help and hear enough frustration in their voice to realize the problem is bigger than the question you called about. You ask for a second opinion and get a completely different way of looking at the situation. Those moments rarely look important while they are happening, but over time they become part of how an organization learns.

Some of them also begin with inefficient little interactions that AI is getting very good at removing.

That is where I think leaders need to pay attention. Most organizations are going to measure AI adoption through the things we already know how to count: hours saved, documents produced, tasks completed faster, fewer meetings, faster turnaround times, more output from the same number of people. Those are legitimate measures, and leaders should know whether the tools they are paying for actually improve the work.

But imagine an employee used to need six hours to complete something and now finishes it in two. I want to know more than the four hours we saved. Maybe that person now has time to think more deeply, spend time with patients or customers, or help somebody else. Maybe the organization simply gives them more work. Maybe, without anyone intending it, they spend more and more of their day working by themselves because they no longer need to ask anybody for help.

Those outcomes would look very different if you were standing beside the person experiencing them. From a distance, they might all look like improved productivity.

People do not learn an organization entirely from its files. They learn who has good judgment. They figure out who remembers the history and who will tell them when an idea is bad. They learn who stays calm when something falls apart, who catches details other people miss, and who will make time when they need help. They also learn each other, slowly, through hundreds of ordinary interactions that nobody would ever call strategic.

If AI removes some of the practical reasons people used to need each other, those relationships will not suddenly disappear. They may simply develop less often. That is harder to notice and harder to measure.

I do not think the answer is more forced interaction. I have no interest in replacing eliminated email with another standing meeting or creating artificial exercises because people are talking less. I think the better answer starts with paying attention to what actually changes when AI becomes part of everyday work.

When an organization rolls these tools out broadly, leaders should ask people what has become easier and what has quietly disappeared. Are newer employees still finding people who will teach them how the place really works? Does information still move between departments? Are people still asking each other what they think after AI has already given them a perfectly reasonable answer? Are managers noticing changes in their teams that would never appear in an AI-generated summary?

Those questions interest me more than another dashboard showing adoption rates.

I believe deeply in what AI will allow people to do. I have watched it give one person capabilities that once required a small team. I have seen people who struggled with writing finally get their thoughts onto the page. I use it myself to understand subjects faster, organize complexity, and get through work that used to take much longer.

I want more of that.

I also know from years of leading people that some of the most important things inside an organization are difficult to count. Trust matters. Judgment matters. History matters. Knowing when somebody is struggling matters. Knowing who to call when the normal process is not enough matters.

AI will make individuals more capable, and that will change the way people rely on one another. Some of those changes will be healthy and overdue. Others might cost us something we do not recognize until much later.

The research does not answer that question yet. I think it gives us a good reason to start paying attention to it now.

*The research referenced here is “Adoption of Generative AI in the Workplace: Increasing and Shifting the Balance of Productivity and Communication Activity,” submitted August 16, 2026\. The preprint examined Microsoft 365 activity across 11 international companies and more than 40,000 Copilot-enabled users. It has not yet been peer reviewed.* [*Read the preprint on arXiv*](https://arxiv.org/abs/2608.15550?ref=mikebakerhq.com)*.*