Everything was working perfectly with AI… until someone asked it to draw a dog.

When AI came — especially tools like Codex, Claude — where it can access all your code directories, create new files, execute code… you just have to command… it was fascinating.

Generating, running input/output just by a few commands.

— -

Let’s take an example

I was given a task to write code to create a beautiful painting automatically.

Earlier, I would be focused on:

• which tools to use
• if code would produce errors
• will it run?
• will I be able to write proper code with proper output on time or not?

But when I used AI… I didn’t have much of these problems.

It had issues fully executing initially, but after a few chats I was able to create such code successfully.

I only know Python, yet it even created a beautiful UI — which was fascinating for me… to see my work without depending on anyone else.

Everything was too easy.

It was good for me. My job done. No efforts. Output is also there. Plus additional outputs are also there.

I ask it to draw a cat — it could draw.

So in that way, it did reduce time, effort… and also made me a bit lazy and comfortable too… because I didn’t have any problem to solve.

— -

But then came another part of it

I was asked to demonstrate the product.

I showed it working.

Then I was asked — what can it draw? I said — anything.

So we asked it to draw a dog.

It said it couldn’t.

There was no error message.

I tried tweaking something… then it drew a pencil sketch dog.

When I checked logs — it said it doesn’t have the right colors for a real dog.

Now the system proved useless.

— -

Pressure started building

I had to improve it somehow.

I went back and tried to read the code… but it was too professional for me. Things were hidden, so I couldn’t understand what was happening.

It was too much pressure to deliver.

Even if I spent 24 hours a day, it became difficult for me to understand things.

So I thought of asking for specific requirements:

• what is the scope?
• what exactly is needed?

And then limit it — rest we’ll do later.

Again I used AI to make changes and include those specific things.

Now this time, things in the scope were getting drawn fine.

— -

But again…

When we tried a simple thing like a pen — it couldn’t.

And I had no honest explanation.

Because I don’t know how it’s working. AI knows.

Uff… it all backfired.

It actually increased my time, effort… everything more.

— -

Taking a step back

Now there was a break.

This time I thought of doing everything by myself.

I was scared to use AI.

But the previous problems came back — code worries, errors, time…

So I had to figure out a way to use AI efficiently.

— -

A different way of thinking

I practice arts too… so that helped me focus with more concentration.

Same problem: code to automatically draw something. But a different approach.

I started thinking — how would I do it normally?

I’ll see if I have the right tools:

• pencil, eraser, colors
• types of colors — crayons, watercolor, poster colors
• canvas

Then I would know:

• how much I have
• what colors
• quantity and quality
• size of canvas

These things help me answer:

• what type of drawing I can draw
• only sketches or a portrait
• how big I can draw
• whether my colors would be enough
• when I need to refill

— -

Mapping it to code

So when I mapped it to code:

• tools → what tools I require
• limits → what are the limits
• usage → which works for what

First I researched on these (with AI).

Then I thought about arrangement: canvas → draw rough sketch → color → let it dry

Based on my requirements, I planned placement of tools.

Then I provided AI with:

• specific tools
• specific structure/workflow
• strict NO to additional things without permission

— -

What changed?

Now still the code was difficult to read…

But the complete workflow was decided by me.

To confirm, I tested connecting points:

• what input I give
• what output I expect

And finally…

AI actually helped me reduce coding time.

— -

So what’s the reality?

AI doesn’t reduce coding time by default.

It reduces time when you know what you’re doing.

Otherwise… it just shifts the problem.

And sometimes… makes it bigger.