I used to think that using AI more often would mean eventually handing over a large part of my work to it.

That hasn’t been my experience.

The more useful change has been much less dramatic.

AI has gradually taken over pieces of work that used to sit between the things I actually wanted to accomplish: summarizing something before I could understand it, turning rough thoughts into a first draft, organizing information, rewriting something for a different audience, or giving me a starting point when I was staring at a blank page.

The work itself didn’t disappear.

The friction around the work did.

And that distinction matters.

There is a lot of discussion about AI replacing jobs, workers or entire departments. But when I look at AI from the perspective of an ordinary workweek, the more immediate question is much smaller:

Which tasks can I stop doing manually?

And just as importantly:

Which tasks should I continue doing myself?

That second question is where things get interesting.

AI didn’t replace my job. It replaced pieces of my process.

This is probably the simplest way I can describe the change.

Before using AI regularly, a task often looked like this:

Think → research → organize → draft → rewrite → edit → finalize

Now some parts can become:

Think → AI assists → review → refine → finalize

That’s not the same as handing the entire task to AI.

The human still decides what needs to be done.

The AI simply reduces some of the repetitive work between the beginning and the end.

And there is growing evidence that this kind of assistance can have a measurable effect.

A well-known NBER study examined 5,179 customer-support agents using a generative-AI assistant. The researchers found that access to the tool increased productivity by 14% on average, with a 34% improvement for novice and lower-skilled workers. The effects were much smaller for the most experienced workers.

That finding is particularly interesting to me because it doesn’t suggest that AI simply makes everyone equally faster.

The benefit depends on the person, the task and how the AI is used.

That’s exactly what I’ve found most useful to think about.

So what did AI actually replace?

Not my judgment.

Not my responsibility.

Not the final decision.

Mostly, it replaced small pieces of repetitive cognitive work.

Here are the areas where I find the biggest difference.

1. The blank page

The blank page is surprisingly expensive.

Not financially.

Mentally.

I can know exactly what I want to say and still spend far too long deciding how to begin.

AI is particularly useful here because I don’t have to treat the first output as the finished product.

I can use it as a starting point.

Give it the basic idea.

Ask for possible structures.

Turn rough notes into an outline.

Generate several ways of approaching the same subject.

Then I take over.

That’s an important distinction.

AI can create the first version without owning the final version.

For me, that changes the psychology of writing.

Instead of asking:

“How do I start?”

I can ask:

“Which of these starting points is actually worth developing?”

Those are very different problems.

2. Summarizing information

This is probably one of the clearest areas where AI has replaced a repetitive step.

Long article.

Long report.

Meeting notes.

Research material.

A collection of messy information.

The old process was essentially:

Read everything → take notes → organize notes → decide what matters.

AI can now help with the middle of that process.

That doesn’t mean I should blindly trust a summary.

It means I can use AI to create a first pass and then go back to the source when accuracy matters.

This distinction becomes especially important with statistics, quotations, research findings and anything that could be misrepresented when condensed.

A summary is useful.

A summary is not automatically a source.

That’s one of the habits I think matters more as AI becomes part of everyday work.

3. Rewriting the same information

Another task AI has made considerably easier is rewriting.

The underlying information may stay exactly the same, but the presentation changes.

A paragraph written for a report may need to become:

• a short email,
• a social post,
• a website introduction,
• a simpler explanation,
• a list,
• a headline,
• or a more concise version.

Before AI, each variation required another round of writing.

Now I can provide the original material and ask for a transformation.

But again, there is a boundary.

I don’t want AI deciding what I believe.

I want it helping me express what I already mean.

That difference keeps the human voice in the process.

4. Organizing messy information

AI is also useful when information exists but isn’t organized.

Imagine receiving twenty messages containing different pieces of information.

Or collecting notes throughout a project.

Or having a document full of ideas that were written in no particular order.

The information isn’t necessarily difficult.

It’s just messy.

AI can help identify:

• themes,
• categories,
• repeated points,
• action items,
• missing information,
• questions that need answers.

That’s a valuable form of assistance because organization is often necessary before meaningful work can begin.

And it doesn’t require AI to make an important decision.

It simply makes the raw material easier to work with.

5. The first-pass routine

There is another category that is easy to underestimate.

The boring first pass.

Checking whether something is complete.

Turning notes into a checklist.

Finding obvious inconsistencies.

Creating a preliminary structure.

Producing a rough classification.

Generating a list of possibilities.

None of these tasks necessarily represents the most valuable part of the work.

But they consume attention.

And attention is limited.

Microsoft’s 2025 Work Trend Index found that 80% of the global workforce reported lacking enough time or energy to get their work done, while employees were interrupted by a meeting, email or notification approximately every two minutes on average.

That makes the appeal of AI assistance easier to understand.

The problem isn’t always that people have too much work.

Sometimes they have too much friction between pieces of work.

What AI hasn’t replaced for me

This is the part I think gets lost in most AI discussions.

There are tasks where I still want a human — and sometimes specifically me — involved.

Not because AI is incapable of producing an answer.

Because producing an answer isn’t always the same thing as knowing what the answer should be.

Here are five things I would still keep firmly on the human side.

1. Deciding what is actually worth doing

AI can give me ten ideas.

That doesn’t mean I should pursue any of them.

It can identify trends.

It can suggest topics.

It can produce possible strategies.

But someone still needs to decide:

Is this worth my time?

That decision depends on context.

Maybe an idea sounds impressive but doesn’t fit the audience.

Maybe something is technically possible but strategically pointless.

Maybe the easiest option isn’t the best option.

AI can help me think through the choices.

I don’t want it making the underlying decision automatically.

The distinction is subtle but important:

AI can expand the menu. I still choose the meal.

2. Final judgment

This is probably the biggest one.

An AI-generated answer can sound extremely confident.

Confidence isn’t evidence.

If I’m publishing an article, making an important recommendation or presenting a statistic, I still want to know where the information came from.

That’s particularly important because AI-generated content can contain incorrect or unsupported claims.

For that reason, I treat AI output as something to evaluate, not something to automatically approve.

This is also why I prefer using primary sources whenever possible.

If a statistic comes from a research paper, I want the research paper.

If a price comes from a company’s official pricing page, I want the official page.

If a claim comes from a government source, I want to check the government source.

AI can help me get there faster.

It shouldn’t remove the need to get there.

3. Original ideas and creative direction

This one is harder to measure.

An AI system can generate hundreds of ideas.

But quantity isn’t the same as originality.

The interesting part of creative work is often the connection between things that don’t obviously belong together.

A person brings:

• experience,
• taste,
• memories,
• preferences,
• cultural context,
• curiosity,
• emotional reactions,
• and personal judgment.

AI can respond to those inputs.

But the reason I want to create something in the first place still comes from the human side of the process.

That’s why I don’t see AI as replacing creativity.

I see it as changing the cost of experimentation.

If an idea can be explored in minutes instead of hours, I’m more willing to test it.

And if the idea doesn’t work, I can move on.

That is a meaningful change.

4. Relationships and communication that actually matter

There are messages I don’t want to outsource.

A quick routine update?

Sure, automation can help.

A standard notification?

Fine.

But an important conversation with a person is different.

When the message itself is part of the relationship, writing it becomes part of the relationship too.

There are times when I want to choose the words myself.

Not because AI couldn’t write a polished message.

Because polished isn’t always the goal.

Sometimes sincerity matters more than efficiency.

5. The final responsibility

This is the biggest boundary of all.

If an AI helps me produce something, I still have to own what I publish or send.

That’s particularly important when the result involves:

• money,
• legal information,
• health information,
• sensitive personal data,
• important business decisions,
• public claims,
• or anything where an error could seriously affect someone.

AI can assist.

It can research.

It can summarize.

It can suggest.

But responsibility doesn’t disappear because a machine generated the first draft.

That’s why I don’t think the most useful question is:

“Can AI do this?”

The better question is:

“Who should be responsible for the result?”

The real split: Implementation vs. judgment

This is the pattern I keep coming back to.

AI is becoming very good at parts of implementation.

Humans still have an important role in judgment.

That doesn’t mean AI can’t reason.

Modern AI systems clearly can perform increasingly sophisticated reasoning and problem-solving tasks.

Microsoft’s 2026 Work Trend Index, based in part on an analysis of more than 100,000 Microsoft 365 Copilot conversations, found that 49% of classified Copilot activity supported cognitive work such as analyzing information, solving problems, evaluating and thinking creatively. Microsoft also reported that 66% of surveyed AI users said AI had allowed them to spend more time on higher-value work.

Those numbers are interesting because they point toward a future where AI isn’t simply a faster typing assistant.

It becomes part of how people work through problems.

But even in that environment, someone still has to define the objective, judge the result and own the consequences.

The productivity numbers are promising — but they need context

This is where I think AI conversations sometimes become misleading.

You will see impressive productivity percentages quoted everywhere.

But a productivity improvement in one controlled study doesn’t mean every worker will suddenly become 14% more productive.

The NBER customer-support study is a good example.

The researchers studied 5,179 agents and found an average productivity improvement of 14%, but the effect varied significantly between workers. Novice and lower-skilled workers benefited much more than highly experienced workers.

That’s an important detail.

AI doesn’t create one universal productivity multiplier.

The result depends on:

the task + the person + the workflow + the quality of the AI assistance.

That’s why I would be suspicious of anyone promising a fixed number of hours saved for everyone.

What changed in my week, then?

If I strip away the hype, the biggest change isn’t that AI “does my work.”

It changes what happens around my work.

I spend less time staring at a blank page.

Less time turning long information into a first summary.

Less time rewriting the same idea for different formats.

Less time organizing messy notes.

Less time doing certain repetitive first passes.

And that creates something more valuable than simply saving a few minutes here and there.

It creates attention.

I can spend more of that attention deciding what matters.

That’s where I think AI becomes genuinely interesting.

But I still want some friction

This may sound strange after an article about saving time.

But I don’t want every part of my work optimized.

Some friction is useful.

Writing something myself can make me think more carefully.

Reading the original source can reveal something a summary missed.

Editing a sentence can change the idea itself.

Talking to a person directly can communicate something that a perfectly optimized message cannot.

The objective shouldn’t be:

Remove every manual task.

It should be:

Remove the manual tasks that don’t deserve my attention.

That’s a much healthier way to approach automation.

The five things I’d keep human

If I had to reduce the entire article to five things, these would be the ones:

1. Choosing what matters

AI can generate options.

I decide which ones deserve attention.

2. Verifying important information

AI can help me find and organize information.

I check important claims against reliable sources.

3. Creative direction

AI can accelerate experimentation.

I decide what I actually want to create.

4. Meaningful communication

AI can improve wording.

I decide when the words should come directly from me.

5. Responsibility

AI can assist with the work.

I remain responsible for the final result.

The most useful question isn’t “What can AI replace?”

It’s:

“What should I stop spending my attention on?”

That’s a much more practical question.

If a task is repetitive, predictable and easy to check, AI may be a good candidate.

If a task requires personal judgment, accountability, emotional intelligence or an understanding of context that isn’t captured in the available information, I would be much more cautious.

And sometimes the answer will be somewhere in between.

AI can prepare the work.

I finish it.

AI can suggest.

I choose.

AI can organize.

I verify.

AI can produce a draft.

I decide whether the draft deserves to exist.

That isn’t AI replacing the human part of work.

It’s AI changing where the human part happens.

What I’d actually try next

If you’re wondering where to begin, don’t start by asking which AI tool you should subscribe to.

Start with your week.

Think about the tasks you repeat.

Which ones make you think:

“I’ve done this before.”

Then look for the parts that are:

• repetitive,
• language-heavy,
• time-consuming,
• relatively easy to check,
• and low-risk if the first attempt isn’t perfect.

Those are usually better candidates than tasks that require your personal judgment from beginning to end.

Try one.

Measure what changes.

Then decide whether it’s worth keeping.

You may discover that AI doesn’t replace nearly as much as the headlines suggest.

But the small amount it does replace may be exactly the part of the process you were happiest to stop doing.

Final thought

I don’t think the future of work is simply:

Humans vs. AI.

The more realistic picture is messier.

Some tasks will be automated.

Some will be accelerated.

Some will change completely.

And some will remain stubbornly human because the value isn’t just in producing an output.

The value is in deciding what the output should mean.

That’s why I’m less interested in asking whether AI can replace a job.

I’d rather ask what happens to the job when AI removes the repetitive parts.

Maybe the person gets more time to think.

Maybe they handle more complex work.

Maybe they create more.

Maybe they simply finish earlier.

The outcome won’t be identical for everyone.

But one thing is becoming increasingly clear:

AI doesn’t have to replace an entire job to change the way that job feels every week.

Sometimes replacing five small tasks is enough.