There was a phase where I started pushing AI into everything.

• Writing with AI.
• Planning with AI.
• Coding with AI.
• Research with AI.

Even thinking through decisions with AI.

At first, it felt like leverage at scale.

Why not use an extraordinary tool for everything?

That seemed rational. But slowly, something strange happened.
The more I tried using AI everywhere… the worse some of my thinking became.

Breaking the Expectation

We tend to assume maximum tool usage means maximum advantage.

Use AI more. Get more value. Simple.

But I started realising that assumption breaks down. Because not everything improves through automation. Some things degrade. Especially when over-optimised.

And thinking is one of them.

The Insight

What backfired wasn’t AI itself. It was my attempt to make it universal.

I was treating AI as the answer to every cognitive task. And that created subtle problems:

• I accepted first answers too quickly
• I explored fewer original paths
• I began outsourcing rough thinking, not just repetitive work

That was the mistake. Because rough thinking, the messy early stage,is often where the best ideas form. And I was bypassing it. Efficiency was starting to eat into originality.

What I Realized

Some tasks should be accelerated.
Others should be wrestled with.
That distinction matters.

AI is exceptional for:

• expanding options
• reducing mechanical effort
• stress-testing ideas

But there are moments where speed is the enemy.
Moments where slowness produces depth.
And I was losing that.

The Bigger Pattern

I think many people are doing something similar.
Using AI not as leverage, but as default cognition.
That feels advanced.
But often it is overdependence disguised as sophistication.
Just because AI can be inserted into every part of work…
doesn’t mean it should be.
That was a hard lesson.

The Reflection

I still use AI heavily.
But with much more restraint.
Because I’ve come to believe this:

Good use of AI is not about putting it everywhere.

It’s knowing where it should stop.
When I tried doing everything with AI, it backfired because I confused amplification with substitution.
And those are very different things.
Some of our best thinking still happens in the parts no tool should touch.