When AI became my reflex

I don’t know about you, but I am using AI more and more to do things I can actually do myself.

Write an email. Pull together a script. Find a better way to say something I already know how to say.

And I have started wondering why.

Am I being efficient? Am I getting lazy? Do I trust AI more than I trust myself? Why am I handing over so much of how I sound in public? And, slightly alarmingly, what is happening to the skills I am using less?

To be clear, I don’t just take whatever it gives me and hit send. I read it. I question it. I change it. Sometimes I bin the whole thing. Whatever goes out still has to feel like me.

But lately I have been asking myself whether that is enough.

I am not asking whether I should use AI. I am definitely going to keep using it. The question is whether I am using it because I have made a choice, or because opening an AI tool has become my first move.

Far out. The world is a wild place right now.

I think the reflex is the problem

I can feel the difference between asking AI for help on a task and asking it before I have thought at all.

One helps me. The other lets me skip the awkward first bit where I stare at the page, work out what I think and make a few bad starts. That bit is slow and mildly annoying. It may also be the bit that keeps the skill alive.

AI makes it very easy to produce more. I am just not convinced that more output always means more capability.

Reviewing is not the same as doing

When AI gives me something good, I feel productive. Sometimes the result is better than what I would have produced on my own. It is easy to take that as evidence that I am getting better too.

But maybe AI got better and I became a better editor.

That still counts. Editing and judgement are real skills. They are just not the same skills as finding the idea, making the connection or building the first version yourself.

I went looking to see whether this discomfort had any basis beyond me overthinking it. A 2025 study from Microsoft Research and Carnegie Mellon University surveyed 319 knowledge workers about 936 real examples of using generative AI at work. Higher confidence in AI was associated with less critical thinking. Higher confidence in their own ability was associated with more.

This was self-reported research, so it is not proof that AI is making our brains shrink. But I keep coming back to what people said their work had become: less gathering and creating, more checking, combining and supervising.

The problem is that checking relies on what you already know. I can only catch a convincing error if I know enough to recognise it. I can only tell that an argument is weak if I have built enough arguments myself. And I can only protect my voice if I know what it sounds like before AI starts writing.

The learning bit makes this extra awkward

Those of us in learning and development and enablement should probably be paying close attention to this.

We tell people that capability grows through practice, retrieval, feedback, reflection and application. We create activities that make learners do something with an idea because we know that seeing a polished answer is not the same as being able to produce one.

Then we open an AI tool and ask it to remove the difficult bit from our own work.

To be fair, not every blank page is sacred. Writing another routine email is not necessarily keeping me sharp. Some work is genuinely low value, and AI can have it.

Where I am stuck is telling the difference between pointless effort and the effort that is actually keeping me good at my job.

A study with 758 consultants by Harvard Business School and Boston Consulting Group helped me think about this. AI improved speed, completion and quality when the task sat within the model’s capabilities. The researchers also found that capability was uneven. AI could be excellent on one task and struggle with another that looked very similar.

So I do not think the answer is less AI. I think it is being much more deliberate about where we use it and which parts of the work we keep.

Then I started thinking about teams

The same messy question shows up at team level, except now there are bigger things involved: client information, learner data, intellectual property, team capability and the way the organisation sounds in the world.

Most teams did not sit down and design how they would use AI. It snuck in task by task.

Someone uses it to draft learner emails. Someone else pastes in feedback and asks for the themes. Another person uses it to create a client scenario or summarise discovery interviews. Each decision can make sense on its own. Put together, they become a way of working that nobody really agreed.

That does not mean the team is reckless. It just means the way of working arrived before the conversation.

A policy helps. It can tell people what is permitted. But teams also need to talk about what good AI-assisted work looks like for them.

Five conversations I think teams need

1. What are we happy to hand over? Take the team’s regular tasks and sort them into three groups: hand over, work alongside and hold close. Formatting a transcript might be easy to hand over. Creating scenario options might be something you do together. The learning diagnosis or core logic behind an experience might stay close. Your categories will be different. That is why the conversation matters.
2. What can AI see? Be specific about which tools are approved and what information can go into them. Client material, learner data, assessment results, internal strategy, proprietary frameworks and unpublished research should not wander into a tool because somebody is in a hurry.
3. What do humans still own? AI can draft, summarise, challenge and analyse. Somebody still needs to understand the work, make the judgement and be willing to put their name to it.
4. How will the work still feel like us? Give the team a shared view of what good looks like. Use approved examples and source material when the access rules allow it. Talk about the language you use, the language you avoid and the things you believe that may not be the average answer.
5. Which skills are we not willing to lose? Some tasks are worth doing without AI occasionally because people need the practice. This is especially important for anyone still building their baseline. Experienced people can draw on years of doing the work themselves. Newer people may never get those repetitions unless the team creates them on purpose.

Keeping it ours

When people talk about keeping AI-generated work “on brand”, it can sound like a tone-of-voice exercise. I think it goes deeper than that.

If a team does not have a clear point of view, AI will give it something that sounds competent and polished. It just may not sound like that team. Or contain much of what makes the team useful.

A clever prompt cannot fix a missing point of view. The people still need to know what they believe, what good looks like, what their clients and learners actually need, and where they are prepared to disagree with the standard answer.

Before AI-assisted work leaves the team, I would ask:

• Does it contain an actual point of view, or does it simply sound professional?
• Can the person submitting it explain and defend the thinking without asking AI again?
• What came from our experience and understanding of this situation?
• Would another team get almost the same result if they used the same prompt?
• Could we still do the important parts if the tool was unavailable or confidently wrong?

Try this with your team

Take ten tasks your team does regularly and put each one under hand over, work alongside or hold close.

Then ask two questions about every task. What information would AI need to see? Which human skill needs to stay live?

You will probably find disagreements quite quickly. One person will see summarising interviews as admin. Another will see it as the moment where the meaning emerges. One person will think a document is harmless to upload. Another will spot the client information in it. One person will hear the team’s voice in an output. Another will hear generic AI language.

Good. Those are exactly the things worth talking about before habits harden into the way the team works.

Where I am landing for now

I do not have a neat answer. To be honest, I am slightly suspicious of neat answers on this topic.

I am not going back to doing everything the old-school way. I also do not want AI to be my automatic first move.

For now, my rule is this. If thinking is the point of the task, I need to start the thinking myself. If the task is mostly administration, I am happy to hand it over. If AI can add range, speed or challenge, I want to bring it in once I know what I am trying to do.

This article came together that way. The question and the discomfort were mine. AI helped me organise the argument, test the structure, find the research and clean up the writing. Then I put some of the mess back, because apparently that is part of the point.

I suspect the balance will keep moving. But I want the choice to remain mine.

And for teams in enablement and L&D, I think the conversation is worth having now: what are we happy to hand over, where do we want AI alongside us, and what are we choosing to hold close?

Sources

Lee and colleagues, The Impact of Generative AI on Critical Thinking, CHI 2025

Dell’Acqua and colleagues, Navigating the Jagged Technological Frontier, 2023

‍