How to Get Better Results from AI: The 5-Step Framework I Use Every Day
I once asked AI a simple question. Not even to do anything. Just a question.
It removed 20% of my codebase.
The website hit production, and the next morning our entire staff couldn't do anything. Production, sales, logistics, accounting, HR. Everything stood still because AI decided to "help" by modifying files I never asked it to touch.
That was the moment I realized AI without supervision is just an expensive way to break things.
The Problem Nobody Talks About
Everyone online is sharing prompts and hacks and shortcuts. "Just ask AI to do X and watch the magic happen." But nobody talks about what happens when the magic doesn't work.
Here's what actually happens:
- You give AI a task
- It gives you something that looks right
- You copy and paste without reading it carefully
- Something breaks and you spend hours figuring out why
Or worse, nothing obviously breaks. The output just has subtle issues. Wrong file paths. Made-up variables. Styling that doesn't match your brand. Facts that sound right but aren't.
I spent three years hitting this wall before I figured out a system that actually works.
TCREI: Five Letters That Changed Everything
After years of trial and error, I developed a simple framework to make sure I give AI the complete context it needs. I call it TCREI.
Task
Define precisely what you want the AI to do. Not vaguely. Precisely.
"Fix my code" is bad. "Refactor the login function in auth.js to handle the case where the session token is expired" is good. The more specific you are here, the less rework you'll do later.
Context
This is where most people fail. AI doesn't know your project, your business rules, your design system, or what you tried last time. You have to tell it.
Context includes things like: existing code it needs to work with, conventions you follow, what you've already tried, business rules it can't violate.
Without the right context, you're asking for a broken product. Every time.
Result Type
Tell AI what the output should look like. Format, length, tone, style.
This is where you say "give me a Python function, not a full script." Or "keep the email under 5 sentences." Or "don't use em dashes."
If you don't dictate this, AI will make its own choices. And they usually won't match what you needed.
Evaluation
We never trust output blindly. This is your gut check. Did it meet the task? Does it fit the context? Is it actually correct?
For code, this means reading every line. Not skimming. Reading. For content, it means checking if it sounds like something you'd actually say. For anything with numbers, it means pulling out a calculator.
Iterate
Based on your evaluation, decide what's next. If the output was close, push back with specific feedback. If it was completely off, rethink your task and context, or just do it yourself.
This is also where the 5-minute rule comes in: if fixing the output manually would take less than 5 minutes, just fix it. Don't go back and forth with AI trying to get perfection.
Why This Works
TCREI transforms AI from a frustrating guessing game into a supervised workflow. Instead of hoping AI reads your mind, you're giving it exactly what it needs upfront and checking the work on the back end.
It's the same thing you'd do with any new hire. You wouldn't hand a junior developer a vague task and expect perfect results. You'd explain the task, give them context, tell them what the deliverable should look like, review their work, and give feedback.
AI is the same. It's capable, but it needs direction.
The Shift
I went from burning through my AI usage with nothing to show for it, to building systems that save me hours every week. The difference wasn't better prompts or a newer model. It was learning to supervise the output instead of trusting it blindly.
AI is a tool, not a replacement for thinking. TCREI is how I make sure the tool actually works.
This is one framework from my full guide, "How to Actually Get Work Done With AI." It covers the supervision mindset, critical reading, when to push back vs fix it yourself, tool comparisons, and a real case study where I automated content creation and took my approval rate from 20% to 80%.
If you want the complete system: thebizspark.gumroad.com/l/ai-guide