I use AI a lot. I use it to research ideas, find better ways of explaining things, summarize information, challenge my thinking, rewrite awkward sentences and occasionally save myself from spending an unreasonable amount of time doing something that a computer can do in seconds. I have also watched it become much better at tasks that, not long ago, required a human being with a particular skill and years of experience. It can write, code, analyze, translate, summarize, brainstorm and produce images. Used properly, it is one of the most useful tools I have encountered in my working life.

That does not mean I want it doing everything.

There is a strange assumption creeping into conversations about AI that if a machine can perform a task, we should probably let it. The logic is understandable. Why spend two hours doing something manually when AI can do it in five minutes? Why write the first draft yourself when a machine can give you one immediately? Why search through hundreds of documents when an AI system can summarize them for you?

Sometimes those are exactly the right questions. But there are other activities where the time spent doing the work is not merely an inefficiency. The process itself is part of the value. The struggle, the uncertainty, the mistakes and even the occasional frustration can be how we learn, develop judgment and discover what we actually think.

The more I use AI, the more I find myself thinking about that distinction. There are things I am very happy to delegate because I genuinely do not care who performs them. There are other things I still want to do myself, even when I know AI could probably do them faster.

I Still Want to Think Before I Ask AI What It Thinks

One of the easiest ways to use AI is also one of the ways I think we should be most careful about using it. You can have a thought, immediately give it to an AI system, and ask whether it is good. You can then ask the system to improve it, challenge it, expand it and turn it into something polished.

The danger is that you can end up outsourcing the development of the idea before you have actually developed it yourself.

I still want to sit with an idea for a while. I want to discover where I stand before I ask a machine to tell me where I should stand. If I am writing about marketing, for example, I would rather first ask myself what I have actually observed in years of working with different companies and clients. I want to remember the campaigns that surprised me, the advice that turned out to be wrong, the meetings where someone changed my mind and the assumptions I no longer believe.

Only after I have done that do I find AI particularly useful.

This concern is not entirely theoretical. In 2026, writing about her decision to avoid AI tools, technology writer Wendy Liu described the value she found in the slow, frustrating process of learning to code before generative AI became capable of producing applications almost instantly. Her argument was not that the old process was efficient. It was that the difficulty itself taught her how to think. She made essentially the same argument about writing: the struggle with language was part of developing the ability to express an idea.

That distinction matters. If every unfinished thought immediately gets handed to an AI system for validation, there is a risk that we become very good at refining ideas without becoming particularly good at having them.

I Don't Want AI to Have My Opinions for Me

AI can produce a remarkably convincing opinion on almost anything. Ask it whether remote work is good or bad, whether companies should return to offices, whether marketing teams should invest in TikTok, whether university is still worth the money or whether employees should use generative AI at work, and you will usually receive a balanced and reasonable answer.

That is precisely the problem.

Real opinions are often less balanced than AI-generated ones because they are formed through experience. They come from things that happened to us, things we witnessed and things we got wrong. A person might strongly believe that a particular management practice does not work because they spent three years working under it. Another person might believe the opposite because it worked brilliantly in their organization.

I do not want an AI system to smooth those differences away.

Some of the most interesting writing comes from people who have actually changed their minds. Someone who once believed that working longer hours demonstrated commitment but eventually realized it was often a sign of poor prioritization has something useful to say. Someone who spent years believing that more data automatically produced better decisions and then discovered that the problem was interpretation has an experience worth sharing.

AI can help articulate that opinion. I do not want it deciding what the opinion should be.

This is also why I find the growing number of people who deliberately limit their AI use interesting. The Washington Post recently profiled students, workers, artists and professionals who were consciously opting out of AI tools. One student explained that she preferred doing her own schoolwork because she saw independent thinking as part of growing up. A worker avoided his company's internal AI chatbot because he was concerned about inaccurate material being passed to foreign partners. Another professional simply found that writing his own emails was faster than explaining to a chatbot what he wanted to say.

These people are not necessarily anti-technology. They are making a distinction between convenience and value.

I Still Want to Write Some Things Badly First

This may sound unnecessarily inefficient, but I think there is value in writing a terrible first draft.

When you write something yourself, particularly something personal or complicated, you discover things while writing. You begin with one idea and realize halfway through that the real idea is somewhere else. You write a paragraph that does not work but contains one sentence that reveals what you were actually trying to say. You remember an example you had forgotten. You notice a contradiction in your own thinking.

If I ask AI to produce a polished article immediately, I can skip much of that discomfort. The result may be more coherent than my first attempt, but I may also lose some of the thinking that happens during the struggle.

A writing instructor at MIT described a similar problem after confronting students who had used AI to produce creative-writing assignments. He found that some AI-generated stories were polished but emotionally hollow, and eventually adopted a clear no-AI policy in his workshop. His argument was not simply about academic honesty. He believed that wrestling with language, making mistakes and trying to express something difficult were part of the intellectual development that writing was supposed to produce.

That is an important distinction because writing is often treated as nothing more than a method for generating a finished document. It is not. Writing can be a way of discovering what you think.

I still want some of my writing to involve that process.

I Don't Want AI Writing Every Message I Send

There is an obvious convenience to asking AI to write every email, message and response. It can make a difficult email sound more diplomatic, turn a rough thought into professional language and remove the anxiety that comes with staring at a blank screen.

But there is a point at which communication stops sounding like communication.

If every difficult conversation is outsourced to AI, we may become less comfortable having difficult conversations ourselves. If every thank-you message is generated, we may lose some of the awkwardness and sincerity that makes an actual thank-you meaningful.

I am particularly wary of this with messages that carry emotional weight. An apology, a condolence message, a note to someone who helped you or a difficult conversation with a colleague should not necessarily be optimized like advertising copy.

There are situations where sounding slightly imperfect is part of being human.

The Washington Post's reporting on people limiting AI use included precisely this kind of resistance. Some people found that AI tools were being inserted into everyday software regardless of whether they had asked for them, from suggested writing to generated images. Their objection was not necessarily that AI was incapable of doing these things. It was that they did not see sufficient value in handing over activities they could comfortably perform themselves.

I understand that position. I would rather send someone a message that sounds like me than a flawless message that could have been sent by anyone.

I Don't Want AI to Choose What I Read

Recommendation algorithms have already changed this part of our lives. Streaming services decide which shows appear in front of us. Social platforms decide which posts we see. Search engines determine which information gets our attention. AI can now go a step further and summarize, filter and personalize information specifically for us.

That sounds wonderful until you consider what gets removed in the process.

Some of the best things I have read were not things I would have searched for deliberately. I came across them because someone recommended them, because I stumbled onto them or because an article happened to be next to something I was already reading. Serendipity is difficult to optimize because by definition you cannot know in advance what you are going to discover.

There is a risk that increasingly intelligent recommendation systems create increasingly narrow information environments. If AI knows what I usually like, it can give me more of it. But sometimes what I need is something I would not have chosen.

I still want the possibility of being surprised.

I Don't Want AI Making Every Decision Easier

One of the most attractive promises of AI is that it can reduce decision fatigue. Give it the options, explain your priorities and ask it to recommend the best choice.

For trivial decisions, that is fantastic. I have no philosophical objection to asking a machine which spreadsheet formula to use or which format makes more sense for a document.

But important decisions are different.

Sometimes the difficulty of a decision is useful. You have to compare competing priorities. You have to acknowledge uncertainty. You have to decide what matters most. You have to accept that there may not be a perfect answer.

If AI always gives us a recommendation, we can become accustomed to treating uncertainty as a technical problem rather than a human condition.

I do not want an algorithm to make every difficult decision easier simply because I find uncertainty uncomfortable. Some decisions should remain mine precisely because I am the person who has to live with the consequences.

I Still Want to Learn Things the Hard Way Sometimes

There is a growing temptation to use AI as an answer machine. Why struggle with something when the explanation is immediately available?

But struggling with something is often how you learn it.

A student who asks AI to solve every difficult mathematics problem may complete more homework while learning less mathematics. A junior developer who asks AI to debug every problem may deliver more code without developing the instincts required to diagnose problems independently. A marketer who asks AI for every campaign idea may have a full content calendar without becoming particularly good at understanding customers.

This is not just an abstract concern. Teachers are already changing how they teach because of it. In Fort Worth, Texas, high-school teacher Chanea Bond responded to widespread access to generative AI by moving her American literature and composition classes toward handwritten work. Students journaled in notebooks and completed most assignments by hand rather than on their school-issued laptops. The point was not nostalgia for paper. It was to make sure students were actually doing the thinking and writing themselves.

There is something important in that decision. The teacher was not claiming that students would never encounter AI. She was protecting a particular learning environment in which the struggle was itself part of the education.

That is a distinction worth keeping in mind at work too.

The frustration is part of the learning process. AI can be an excellent tutor, coach and explainer. The difference is whether I am using it to help me understand something or using it to avoid understanding it.

I want to preserve some of the struggle.

I Don't Want AI to Replace Human Expertise With Confidence

One of the most impressive and worrying things about generative AI is how confidently it can communicate.

A human expert might say, "I'm not completely sure, but based on these circumstances, I think this is probably the answer." An AI system can produce five paragraphs of beautifully structured reasoning that sound much more certain.

That creates a dangerous psychological effect. We naturally associate fluency with competence.

But fluent language is not the same as knowledge.

This is one reason I found the experience of Steve Royle, a professor of cell biology at the University of Warwick, particularly interesting. In a Guardian article about people deliberately limiting AI use, Royle explained that he uses ChatGPT for some of the "grunt work" of coding but does not want it generating code from scratch because he believes he can spend more time debugging an overambitious AI solution than he would have spent writing the code himself. He also expressed concern that excessive dependence could cause his own coding skills to deteriorate.

That is a much more interesting position than simply saying AI is good or bad.

Royle has found a useful boundary. He is happy to let AI handle a particular category of tedious work, but he does not want to surrender the expertise required to understand and control the work.

That is probably the balance I want too.

I Don't Want AI to Do the Parts of Work That Give Work Meaning

Not everyone experiences work as meaningful, of course, and plenty of jobs contain large amounts of unnecessary administration. Removing pointless tasks is one of the best uses of AI.

But some work is satisfying precisely because you are doing it.

A designer might enjoy solving a visual problem rather than simply selecting the best AI-generated option. A writer might enjoy finding the right sentence. A programmer might enjoy figuring out why a system is failing. A teacher might enjoy discovering the explanation that finally makes something click for a student.

If we optimize every profession around maximum efficiency, we risk removing not only the tedious parts but also the parts people actually enjoy.

There is some evidence that people feel this distinction even when they are perfectly comfortable with AI as a tool. A 2025 survey published by Scientific American asked people about AI-generated art and found that most respondents were uncomfortable with the idea of AI simply creating art on its own. At the same time, many were comfortable with AI being used as a tool when a human artist remained heavily involved in guiding the creative process.

That distinction is fascinating because it is not really about whether the machine is capable of producing an attractive image, piece of music or written passage. It is about authorship and human involvement.

People appear to care about who is behind the work.

That does not mean we should reject automation. It means we should be more thoughtful about what we automate.

I Don't Want AI to Become My Substitute for Experience

This may be the most important limitation I want to maintain.

There are things I can ask AI about that it knows far more about than I do. There are also things where my own experience is the asset.

After years of working with different organizations, clients, managers, teams and campaigns, I have accumulated a collection of patterns. Some are difficult to explain. Sometimes you simply recognize that something is going wrong before you can articulate exactly why.

That kind of judgment comes from exposure.

AI can tell me what the textbook says. It can summarize what hundreds of experts have written. It can even challenge my assumptions. But I do not want it to replace the process through which I develop my own judgment.

Experience is slow to acquire, and that is part of its value.

This is particularly relevant in professions where AI can produce convincing first drafts. A junior marketer can ask AI for a campaign strategy and receive something that looks remarkably sophisticated. But an experienced marketer may recognize that the strategy is based on an incorrect understanding of the customer, that the proposed positioning is indistinguishable from competitors or that the measurement plan is designed around metrics that look impressive but do not actually answer the business question.

The machine can produce the document. Experience tells you whether the document deserves to be believed.

I Don't Want AI to Replace the Human Relationship

There is one category of work I would be especially reluctant to outsource completely: anything where the relationship is itself the product.

A therapist can use technology to organize notes. A doctor can use AI to summarize medical information. A teacher can use it to prepare exercises. A manager can use it to draft administrative documents.

But none of those things means the human relationship is unnecessary.

A student does not simply need information. They need encouragement, challenge and someone who notices when they are struggling. A patient does not simply need a correct explanation. They need someone who understands their fear and circumstances. An employee does not only need instructions. They need a manager who understands them well enough to know when they need support and when they need to be challenged.

Recent debates about AI in education illustrate this tension. Critics of AI tutoring have argued that learning is not simply the delivery of personalized information and that human interaction plays an important role in developing judgment, creativity and social skills. Even advocates of educational AI generally acknowledge that the technology needs to be integrated carefully rather than treated as a universal replacement for human teaching.

There are some things we do for people rather than merely to them.

I do not want AI to erase that distinction.

I Don't Want AI to Decide What Counts as Creative

This is becoming increasingly complicated because AI can now produce things that look creative.

It can generate an illustration in seconds. It can produce a song in a particular style. It can write a short story. It can develop a visual identity or suggest hundreds of advertising concepts.

But creative work has never been only about the finished object.

A novel is also the years someone spent observing people before writing it. A photograph is also the photographer's decision to stand in one particular place at one particular time. A painting can carry the history of the person who made it. A marketing campaign can reflect everything a team has learned about a particular audience.

This is why the debate around AI-generated art is more complicated than whether the output looks good. The Scientific American survey found that people were considerably more comfortable with AI assisting a human artist than with AI acting as the artist itself. Many respondents wanted significant human guidance to remain part of the process.

That feels intuitively right to me.

I do not need every creative output to be entirely human-made. But I want to know that there is a human somewhere who had something they genuinely wanted to express.

I Don't Want AI to Do Everything Just Because It Can

The most important word in this entire conversation may be "want."

We have spent decades asking technology what it can do. Can computers calculate faster? Can the internet connect people instantly? Can smartphones put almost every piece of information in our pockets? Can AI write an article in ten seconds?

The answer to all of these questions is increasingly yes.

But capability does not create obligation.

Just because AI can write my email does not mean I should let it. Just because it can produce an illustration does not mean an illustrator should stop drawing. Just because it can answer a student's question does not mean a teacher should disappear. Just because it can generate a first draft does not mean I should stop thinking.

This is why I find the people deliberately limiting AI interesting. They are not necessarily rejecting technology. Many of them use it selectively. They are simply drawing boundaries around activities they believe are worth doing themselves.

The boundaries will be different for everyone.

The Point Isn't to Keep AI Out

None of this means I want to work without AI. Quite the opposite. I expect AI to become increasingly integrated into how I work, and I will probably become more dependent on it for certain tasks as the technology improves.

The important distinction is between using AI as a tool and allowing it to become the default intermediary between me and everything I do.

I want AI to remove friction where friction is pointless. I want it to handle repetitive tasks, help me research, catch mistakes, organize information, explore possibilities and challenge ideas. I want to use it for the things where speed genuinely improves the outcome.

But I also want to retain some activities precisely because they are mine.

I want to think before I ask. I want to write before I generate. I want to make some decisions without requesting a recommendation. I want to struggle with difficult ideas occasionally. I want to form opinions from experience. I want to remember things imperfectly. I want to have conversations that sound like conversations rather than perfectly optimized pieces of communication.

There is a bigger question underneath all of this. If AI eventually becomes capable of doing almost everything faster than we can, what exactly are we trying to optimize for?

If the answer is simply efficiency, the logical conclusion is uncomfortable. We could eventually build systems that remove enormous amounts of human involvement from everyday life. But a life optimized entirely for efficiency may not be the same thing as a life optimized for meaning, curiosity, connection or satisfaction.

I do not think the answer is to reject AI. That would be as unrealistic as refusing to use computers because they changed office work. The better answer is to become more deliberate about what we hand over.

Some tasks are worth delegating because the human contribution adds little. Other activities are valuable because a human being is doing them, thinking through them, struggling with them or experiencing them.

AI will probably become capable of doing many more things than I want it to do for me.

And perhaps that is the point. The challenge of the AI era will not simply be learning what machines can do. It will be deciding what we still want to do ourselves.