AI Side Quests: where do you safely experiment with AI?

I’ve been thinking a lot about where AI experimentation actually belongs in an engineer’s workflow.

Using AI everywhere feels risky. Avoiding it entirely feels like leaving learning on the table.

What’s been working for me is separating production work from what I call AI Side Quests.

An AI Side Quest is a small, low-stakes project where:

• scope is tiny
• time is capped (a few hours, maybe a day)
• quality doesn’t matter
• AI is used aggressively

The goal isn’t shipping or best practices — it’s building intuition fast.

Side quests are where I intentionally break that rule — safely, without long-term consequences.

It’s helped me:

• reduce AI skepticism
• move faster when exploring new tools or languages
• keep my fundamentals intact
• rediscover the fun of building without pressure

Curious how others handle this:

Do you separate “experimentation” from “real work”?
Where do you draw the line with AI usage?
Have you found a safe way to explore without polluting production code?

Would love to hear how others are navigating this.