Calibrating AI Before It Smothers You Out

For a while I was avoiding AI tools. I thought its a trend, which like most things, would fade away.

I told myself it was about skills. About not outsourcing the thinking. About staying sharp. There was a certain satisfaction in working through something the long way and arriving at an answer on my own.

That phase lasted until I was deep in a project and the people around me were shipping faster, unblocking themselves quicker, and not grinding through the same friction I was choosing to grind through. I gave in. I started using it.

And then, slowly, I went too far the other way before I knew it.

What I started noticing

It crept up on me. I would open a chat window before I had even sat with the problem for five minutes. I would ask it to structure my thinking before I had formed any thoughts of my own. I would get a plan, feel good about the plan, and execute something I had not really designed.

The output was often fine. Sometimes it was good. But I started noticing that when I was in a conversation about something I had built, I was slightly less sure of my own reasoning. When someone pushed back, I would second-guess faster. When I was offline or in a meeting or just needed to think something through, I felt slower.

The confidence I used to have in my own ability to work through a hard problem had quietly eroded. I had been handing that process off without noticing. It got to the point that I ran even a Teams reply through an AI agent to make it sound smart.

That is the part that got to me. Not the code quality, not the output. The fact that I was losing trust in my own ability to think.

What I have decided now

I am not going back to avoiding it entirely. That was also a mistake. These tools are useful, and refusing to use them is not discipline, it is stubbornness dressed up as principle. It also means I am missing out on learning how to use them effectively, which is a skill in itself.

But I am trying to draw a sharper line now.

Thinking stays with me. If I do not understand the problem, if I have not formed a view on the right approach, if I cannot articulate the tradeoffs, I am not touching an AI yet. Sit with it. Think. Write something down. Let the discomfort do its job. Let the old feeling of struggling through something come back.

AI earns its place on the execution side. Once I know what I want, once I have reasoned about it, once I have a clear intent, then I will use it to move faster. Boilerplate, scaffolding, repetitive transforms, looking up syntax I know conceptually but cannot recall exactly, writing the boring middle sections of something whose structure I already own.

Mundane cognitive work is fair game. Summarising something long that I need to skim. Generating test cases for something I have already designed. Turning rough notes into something readable. This is fine. This does not hollow anything out.

The rule I am trying to follow: if using AI here means I would not know how to explain or defend this later, I should not use it yet.

Why this matters more in engineering than other fields

In most knowledge work, the end product is what people see. But in engineering, especially in a team setting, you are not just producing an output. You are expected to understand it, maintain it, extend it, and explain it. You need to be the person your teammates can rely on to actually know what is going on.

Confidence in your own judgement is not a soft skill. It is load-bearing.

When you outsource your thinking too often, you do not just lose the skill, you lose the awareness that you have lost it. You start feeling capable because things are shipping, without noticing that you would struggle to do it without the crutch.

Where I am now

I still use AI every day. It would be wasteful and utterly foolish not to. But I am more deliberate about when I reach for it and what I am asking it to do.

I try to arrive at the conversation with an opinion, not a blank slate. I use it to move faster on things I have already thought through, not to replace the thinking. I push back on outputs more than I used to, because I have usually spent enough time with the problem to know when something is off.

The goal is not to be the person who avoids the tools. The goal is to be the person who still knows what they are doing when the tools are not there. The goal is simply to learn from my mistakes like I did before.

That is the version I am trying to get back to.