The kooky world of AI

AI has a particular talent for being both impressive and ridiculous in the same sentence.

One moment it is helping you untangle a gnarly bug. The next moment it is confidently inventing a library that does not exist.

I am choosing to enjoy the weirdness while still treating it like a power tool.

AI is not magic. It is a compressor

A decent mental model for me is: AI compresses patterns from a lot of human writing and code, then predicts what should come next.

That can be incredibly useful.

It can also produce answers that sound right and are still wrong.

So I try to keep one rule in mind:

If it matters, verify it.

The best prompts are basically good briefs

When AI output is bad, my first instinct used to be “the model is bad.”

Now I usually assume my prompt is vague.

What helps:

• A clear goal (what “done” looks like)
• Constraints (stack, performance, accessibility, tone)
• Inputs (existing code, a link, the exact error)
• A request for tradeoffs (options, pros/cons)

That is just good communication. AI forces the issue.

It is a mirror for your own thinking

When I ask an AI to explain something and it gives me mush, it usually means I do not understand the thing as well as I thought.

That is not a failure. It is a clue.

Sometimes the most valuable result is not the answer, but the realization that I need to refine the question.

The weirdest part: it can be fun

There is a playful side to AI that I do not want to lose.

Sometimes I use it like a creative partner:

• “Give me 10 silly metaphors for state management.”
• “Rewrite this paragraph like a seaside postcard.”
• “Invent a fake town in Mayo and describe its annual festival.”

Half of it is nonsense. The other half shakes something loose.

My current stance

• Use it to accelerate drafts, debugging, and exploration.
• Keep responsibility on the human side of the keyboard.
• Treat output as a starting point, not a verdict.

If you are using AI in your work, what is your most reliable use case so far?