I treat my AI like a junior assistant who’s eager to please but needs me to...

I treat my AI like a junior assistant who’s eager to please but needs me to spell things out.

I’ve learned the hard way with real assistants: if I’m vague, like saying “grab some stats” without details, they’ll flub it, especially if it’s my niche and they’re clueless. Same deal with AI: if my instructions aren’t razor-sharp, I get junk results.

And AI’s got quirks, it daydreams, chasing random tangents; it’s got a dyslexia-like knack for jumbling numbers or quotes; and it’s such a people-pleaser, nodding along when I’m wrong instead of saying, “Uh, boss, you goofed.”

Plus, it’s like it’s been on Reddit too much, with opinions that sound like they came from super left-wing parents, so I watch for that bias.

To make it work for me, I feed it pre-made documents with exactly what I need and use ChatGPT’s memory feature to keep it in line. Even then, I gotta refresh those instructions often, like redirecting an intern who’s zoning out. This setup’s been a total win for me, turning my shaky AI into a solid helper.