The prompt I’m never showing you.

_{Day 43/100}

Hey—It's Tim. 

I spent the weekend staring at a screenshot of my own sins: a single Penny prompt with 2,000+ tokens of “helpful” inputs.

That’s the same as handing a writer 900 words of brief for one H2 section and saying, “go on then.” That writer was Penny.
That writer was me.
No wonder it creaked.

How it went wrong

I kept shoving more into the prompt as models improved.

More rules.
More railings.
More vibes.

The intention was clarity; the effect was fog. We writers don’t need everything at once - we need the right thing at the right time.

Back to the drawing board (literally)

So I did what I always do: pulled out the tablet and drew the whole flow. Boxes, arrows, insults.

I asked, “When does this input actually matter?” Not “can it be included?”

When should it shape the writing? This is my north star for this sequence.

Built like a writer, not a robot

Here’s what actually happens when I hit run—no hype, just the workflow that finally stopped me from drowning Penny in context.

• Template pass. Skim the core brief and sketch the bones. Think index cards: what’s the job of this section, what belongs where, what can wait.
  

• Draft pass. Penny writes a rough draft. Then she literally asks for what she’s missing - examples, receipts, counter-angles - like a good junior who isn’t afraid to say “I need X to finish this.”
  

• Editor pass. We put on the mean hat. Cut flab, tighten claims, add “according to who?” moments, fix rhythm. It’s the markup you’d leave a friend in Google Docs.
  

• Rewrite pass. One clean rewrite in my voice using those notes - and only now tuck in the exact phrases we need for search so nothing reads like a keyword hostage.

It’s just better.

There are 100 and 1 benefits for doing it this way. Some of my favs -

Because my brain (and Penny’s) stops juggling. We only ask one small question at a time, so the answers are better.

• It feels calmer. I don’t open a 2,000-token hairball anymore. I open four small, knowable steps.

• It’s easier to tweak. If the draft over-explains, I nudge the editor pass. I don’t have to surgery a monster prompt and pray.

• It reads human. The final section sounds like me after a coffee—clear, a bit cheeky, with receipts where they matter, not wallpapered everywhere.

• It’s cheaper and faster. Weirdly, the split-up version runs quicker and costs less than half. Smaller thoughts think faster.

I’ll show before/afters tomorrow. We’re wiring up the production flow today; the manual tests already slap.

The sketchbook I wasn’t going to show you

Before a single line of code, I fill a few pages with blocky handwriting, arrows, and mean little notes to myself.

Can I be honest with you?
I’ve been calling this sequence the most important prompt the world will never see - but will know. 

Slightly delusional? Maybe.

But I honestly think if we get this right, the internet gets a better. More, amazing content, written from your experiences. I don’t see a downside.

See ya tomorrow

✌️ Tim "We have prompts at home" Hanson
CMO @Penfriend.ai

Same brain, different platforms: X, Threads, LinkedIn.

P.S. Legit question, would you want to see the different stages of the writing process?
I’ve been debating how we add this in, because I think it’s cool af, but it’s also a messsssssss to see it all. Reply with “show me the working” is you would want to see it.
More than 20, I’ll consider it. Or make it an easter egg for those who said yes and buy.

You gotta buy it to see it. Duh.

Penfriend.ai
Made by content marketers. Used by better ones.