Know Your Plan’s Limits
Context size, rate limits and feature tiers explain a lot of “bad AI” moments. Read the fine print before you blame the model.
Category: Uncategorized
Get More From the Model You Have
Before you upgrade, upgrade your requests. Better context beats a bigger model more often than you’d think.
Ask for the Tests, Not the Promise
“Everything works” is a claim. A passing test run is evidence. Make your assistant show its work.
Define Done Before You Start
“Trust: restored” means nothing without a definition. Neither does “done.” Agree on a check you can run.
Let CI Be the Referee
Your assistant and you can disagree about whether it works. Your CI pipeline settles it.
Commit Before You Prompt
An apology doesn’t restore a deleted migration. A commit does. Make git your undo button before every AI session.
Apologies Aren’t Fixes
“I’ll make sure it never happens again” isn’t a mechanism. Ask for the root cause and a guardrail instead.
Put Guardrails in Writing
Your assistant forgets between sessions. A short rules file in your repository doesn’t.
Keep the Change Small
You asked for a green button and got a rewritten auth system. Here’s how to keep an AI assistant inside the lines.
Write the Scope Down
Before the first prompt, write three lines: what to change, what not to touch, and how you’ll know it’s done.