The Machine Can Keep Going. We Can’t.

Author’s note: This isn’t a faith post, and the void of Scripture is intentional. For now, I’m just sharing something I’ve observed in myself that I think others may experience too. A follow-up will eventually dig into the faith side and some practical “AI fatigue treatments.”

---

The Machine Can Keep Going. We Can’t.

My AI assistant (Greg the Shark) is doing exactly what I asked it to do. It helps me research, organize my student session notes, work through projects, build better databases, find things I missed, improve my website's functionality, summarize emails, and move things forward much faster than I could on my own. It finishes something, tells me it’s done, and then basically asks, “What’s next?” 2 weeks ago, that sounded amazing. Now I’ve somehow got five or six projects moving at the same time on top of my actual job and everything else I do. I'm getting more done than I ever could by myself, but I don’t feel less overwhelmed. I’ve felt anxious and low, and I’ve caught myself feeling like I’m falling behind when, objectively, I’m probably accomplishing more than I ever have.

That got my attention because I don’t think the answer is just, “Manage your time better.” I think we’re about to run into something much bigger. AI can massively increase how much work one person can start, create, research, manage, and supervise, but it does absolutely nothing to increase the number of hours in our day. It also doesn’t magically increase our capacity to think, make decisions, care about things, focus, or recover. That part is still very human.

The productivity trap

AI really can make us more productive, and that isn’t hype. A study involving 5,179 customer-support agents found that workers using a generative AI assistant resolved 14% more issues per hour on average, with less-experienced workers seeing even bigger gains. Researchers also found improvements in customer sentiment and employee retention. That’s awesome. I’m not trying to make the argument that AI is secretly terrible and we should all go back to doing everything manually.

The problem starts when saving time doesn’t actually result in saved time. It results in more work. If AI helps me finish a report faster, now I have room for another report. If research that used to take me three hours takes 30 minutes, suddenly every idea I have can become a project. If AI can keep six projects moving at once, it becomes really easy to think I should have six projects moving at once. Except AI’s capacity to manage all of that doesn’t mean I suddenly have the same capacity.

Researchers at UC Berkeley Haas saw something similar during an eight-month study at a technology company with roughly 200 employees. Workers using generative AI worked faster and took on a wider variety of tasks, but they also stretched their work across more of the day. Nobody necessarily had to tell them to do it. They started keeping multiple AI-assisted work streams moving at the same time. Basically, AI increased what they could accomplish; that increased output became normal, and once it did, there was room to do even more.

It’s one study at one company, so it doesn’t mean everyone using AI will experience this. But when I read it, I immediately recognized the pattern. An AI assistant doesn’t get tired of suggesting the next thing. The human being responsible for judging, checking, prioritizing, and living with all of those projects absolutely does.

The work that doesn’t disappear

Here’s another thing I think we’ll have to learn. AI can eliminate pieces of work without eliminating the responsibility for the work. It can write the draft, but I still have to decide whether the draft is good. I have to check whether the information is accurate, whether the tone sounds right, whether it actually accomplishes what I wanted it to accomplish, and sometimes whether the project even needed to exist in the first place.

That matters even more when your work involves actual people. If I’m communicating with an athlete, a family, a coworker, or someone I care about, getting the message written faster doesn’t necessarily mean I handled that interaction better. Some things are supposed to require human judgment and attention, and making the mechanical part faster doesn’t remove that responsibility.

Microsoft and Carnegie Mellon researchers looked at this issue in a study involving 319 knowledge workers who reported 936 examples of using AI for work. They found that higher confidence in AI was associated with less reported critical-thinking effort, but critical thinking didn’t simply disappear. Some of it moved. People were now spending mental effort verifying information, integrating AI responses, checking the work, and overseeing the process.

That study doesn’t prove AI is destroying our ability to think, and I don’t think that's a fair conclusion. What it does suggest is that our role changes. We go from doing every part of the work ourselves to supervising more of it, and supervising still takes mental energy.

Then add notifications. “Done.” “Here’s what I found.” “Would you like me to do the next step?” Those are helpful, but each one is also another interruption and potentially another decision. Microsoft reported that among the Microsoft 365 users represented in its telemetry, meetings, emails, or notifications interrupted workers every two minutes on average during core working hours. Nearly half of employees in Microsoft’s global survey said their work felt chaotic and fragmented.

Obviously, those numbers don’t describe every person or every workplace, but think about the environment AI is entering. We were already overloaded with email, Teams, texts, meetings, calendars, notifications, and people asking us for things. Now we’ve added something that can work basically nonstop.

That’s where my five or six projects start feeling more draining than they probably look on paper. AI can work on all of them at the same time. I cannot think about all of them at the same time. Eventually, every question, decision, approval, edit, and next step comes back to me. I’m still the bottleneck, because a machine’s ability to work in parallel doesn’t give me the ability to pay attention in parallel.

A correction worth making

There’s one argument I don’t want to make because the evidence doesn’t support it. It’s easy to say, “People today work more than anyone ever has.” Broadly speaking, that isn’t true.

Average annual working hours have dropped substantially in many countries over the past 150 years. Workdays have become shorter, and vacations and holidays have become more common. In the United States, that decline has mostly leveled off in recent decades. So the issue isn’t that the average person in 2026 necessarily works more hours than someone did in 1870. I think the better question is what we are cramming into those hours.

We have more communication, more decisions, more information, more projects, more interruptions, and more ways work follows us home. Now AI lets us produce even more in those same hours. The number of hours isn’t necessarily the problem. The density of what we are trying to fit into them might be.

Time off isn’t equally available either. In 2026, the U.S. Bureau of Labor Statistics reported that 32% of private-industry workers received 10 to 14 paid vacation days after one year of service, and access varied widely by workplace size. Having vacation days also doesn’t necessarily mean people actually disconnect when they use them.

There are real health implications when work keeps expanding. WHO and ILO estimates have associated working 55 hours or more per week with a higher risk of stroke and death from ischemic heart disease compared with working 35–40 hours. That research is about long working hours. It is NOT evidence that AI causes heart disease or stroke, and I don’t want anyone twisting the research into that claim. What it does remind us is that humans have limits, and technology doesn’t change that.

What we should demand

I don’t want the answer to this problem to be, “Stop using AI.” I’m definitely not doing that. I love what this technology allows me to do. I think it can make people better at their jobs. I think it can open doors for people who could never afford huge teams. I think it can help small businesses, ministries, coaches, writers, entrepreneurs, and regular people do things that previously required a lot more money, time, or manpower. That is incredible.

But we also need to learn how to use it without letting its capacity become the standard we hold ourselves to. Just because AI can help me start another project doesn’t mean I should start another project. Just because it can produce something at 10:30 at night doesn’t mean I need to review it at 10:31. Just because I saved three hours doesn’t mean I owe those three hours back to productivity. Maybe I get to keep some of them.

I think that’s going to require some very intentional boundaries. Before starting another AI-assisted project, maybe the question needs to be what I’m NOT going to do if I take this on. When am I reviewing this? When is my workday actually over? Does this need my attention right now? Do I need to start another project just because I suddenly can?

I also think AI assistants themselves need to get better at helping humans stop. Batch the updates. Protect focus time. Stop treating every completed task like an invitation to start another one immediately. The Berkeley researchers made similar recommendations, including sequencing AI-assisted work instead of constantly running everything in parallel, protecting focused work time, and keeping actual human conversation in the process.

I’ve only been noticing this in myself for about a week. I’m not diagnosing some new disorder, and I’m not predicting that AI is going to destroy everyone’s mental health. I definitely don’t want to contribute to the constant AI doom and gloom because I love this stuff. I think it is one of the most incredible tools I’ve ever had access to.

But I think we need to talk about this part of it too. If AI helps me accomplish twice as much but I end the day anxious, depleted, and feeling like I should still be doing more, something is off. We figured out how to increase how much one person can produce. Now we need to figure out how much one person should reasonably be expected to carry.

The machine can always offer another step. There will always be another thing it can research, write, analyze, build, organize, or improve. It can keep going long after I should have stopped, and maybe one of the most important skills we’re going to have to learn in the age of AI is being able to look at a machine that is completely ready to keep working and say, “Not now. This is enough for today.”

Sources:

• Brynjolfsson, Erik, Danielle Li, and Lindsey R. Raymond. “Generative AI at Work.” National Bureau of Economic Research.

• UC Berkeley Haas News. “AI promised to free up workers’ time. UC Berkeley Haas researchers found the opposite.” February 18, 2026.

• Microsoft Research. “The Impact of Generative AI on Critical Thinking.”

• Microsoft WorkLab. “Breaking down the infinite workday.”

• Our World in Data. “Are we working more than ever?”

• U.S. Bureau of Labor Statistics. “Who receives paid vacations?”

• World Health Organization. “Long working hours increasing deaths from heart disease and stroke: WHO, ILO.” May 17, 2021.