A WORKING TOOL FROM DR. JEANENE SINGLETON
Stop Starting Over With AI
Set up one recurring task so you don't have to keep re-explaining it to ChatGPT or Claude.
About 20 minutes. Use it on real work today.
You already know AI can help. That's not the problem.
The problem is that every time the same work comes around, you start over. You explain the context again. You upload the same file. You restate the rules. You dig through old chats looking for the conversation where it finally understood what you wanted.
You're using AI to save time, and using AI is still taking too much of it.
If the task repeats, your setup shouldn't start over.
You're going to pick one recurring task, write down the parts you keep re-explaining, and create a setup you can use again.
FIRST, HERE'S WHAT YOU'RE BUILDING
A real setup from my own work.
My email piles up all day, and the hard part was never writing the reply. It was deciding what mattered.
The job
Review email that came in since the last check and prepare what needs my attention.
What it gets
New messages since the previous review.
What it needs to know
Not every email deserves the same amount of attention. Newsletters, automated notifications, and routine inbox noise shouldn't compete with messages that actually need me.
What it gives back
Meaningful messages sorted into four groups: urgent, needs my response, informational, and waiting on someone else. It can also prepare draft replies when a reply is actually needed.
What good looks like
The genuinely urgent things are visible. Everything else isn't dressed up to look urgent. Drafts sound like me, not like a customer service template.
Rules + boundaries
It can draft. It never sends. If it isn't sure, it says so instead of guessing.
That's the foundation. It's not a clever prompt. It's the job, written down clearly enough that AI doesn't have to start from zero every time.
STEP ONE
Pick the task.
Not your hardest problem. Pick one recurring task with a result you recognize.
Have I done this before, or do I know I'll do it again?
Can I tell a good result from a bad one?
Is there part of this work AI could help prepare, organize, analyze, research, or draft without making the final call for me?
Bonus if you've already used AI for it before. You'll know exactly what you're tired of repeating.
You don't need the perfect task. You need a real one.
STEP TWO
Catch what you keep repeating.
Don't design an AI system. Start with the annoyance.
I always have to tell AI that...
What comes next? The background. Who it's for. What you've already decided. Your preferences. Your process. The constraints. Something obvious to you that AI keeps missing.
Too long. Wrong tone. Missed the point. Made an assumption. Gave you five options when you wanted one.
STEP THREE
Say what good looks like.
You've probably been carrying the standard in your head. Make it visible.
Think about what should come back to you, how it should be organized, what level of detail matters, what it should prioritize, and what it should leave out.
One real example can teach more than another paragraph of instructions.
STEP FOUR
Mark what it doesn't decide.
This is where you draw the line.
Mine can draft an email. It can never send one.
STEP FIVE
Build your setup.
Put what you just wrote into this structure.
THE JOB [What you want help with repeatedly.] WHAT IT GETS [What you'll normally give it when the task shows up.] WHAT IT NEEDS TO KNOW [The context, preferences, past decisions, process, and information you keep explaining.] WHAT IT GIVES BACK [What should come back to you.] WHAT GOOD LOOKS LIKE [Your standards. What makes a useful result useful.] RULES + BOUNDARIES [Always. Never. Stop and ask when.] EXAMPLE [A strong past result, if you have one.]
Save your completed setup somewhere you can use it again. Put it into a ChatGPT Project or Claude Project, or save it somewhere you can paste it in when the work comes back.
Then use it on the next real instance of the task.
If that work is in front of you today, use it today.
STEP SIX
When it misses, fix the setup — not the chat.
When AI gets something wrong, the instinct is to correct it right there in the conversation.
It fixes the answer. You move on. Then next week, you give the same correction again.
Don't repeat the correction. Add it to the setup.
If it made an assumption it shouldn't have, maybe you need another rule. If the tone was wrong, maybe your standard needs to be clearer. If it didn't know something important, maybe that context was still living only in your head.
If it reaches something it shouldn't decide, add a boundary that tells it to stop and ask.
Every miss gives you information.
The goal isn't to make AI know everything. It's to make the setup better instead of rebuilding it.
The first version isn't the final version.
ONE MORE QUESTION
If this worked every time this task came around, what would you want it to give back to you?
Maybe it's time. Maybe it's attention for work that actually needs you. Maybe it's a morning that doesn't start behind. Maybe it's room for something outside work entirely.
Decide now. Otherwise something else will quietly fill the space.
WHAT'S NEXT
The Apprenticeship
You just created the foundation for one recurring task. The Apprenticeship shows you how to turn that foundation into a working apprentice that can handle real work, learn from misses, and improve without starting over.
A chat answers you. An apprentice learns the job.
Available September 16. You're already on the first-access list.