It was past midnight and the terminal was still moving.
I had been building for weeks.
The work was real, the output was enormous, and I sat there feeling like I was somewhere behind an invisible line I could not see.
You know the feeling.
You close a laptop after a long stretch of real work, and instead of "I did a lot today," the quiet voice says "everyone else is further ahead."
I want to tell you what two months of that actually looked like, with the receipts, because the receipts surprised me.
What the two months actually held
Roughly two months. One flat subscription. No metered API bill.
Here is what ran through it, end to end.
- 20.7 billion tokens processed
- 46 million tokens actually written back into files
- 95.6 percent served straight from prompt cache
- 5 models orchestrated across the work, with Opus 4.8 in the lead
- 16,297 dollars of equivalent Claude API value, at public per-token rates
I did not set out to hit a number.
I set out to build, and the number is what fell out the back.
The part that stopped me was not the 16 thousand.
It was the 46 million.
Out of 20.7 billion tokens moving through the machine, only 46 million were words that actually stuck to disk.
That is the ratio of real building.
You churn through an ocean of reading, planning, and dead ends, and a thin river of it survives as the thing that ships.
The number I was proud of is not the one that mattered
The cache did the heavy lifting.
95.6 percent of everything came back from cache, never a fresh charge.
If you have ever felt guilty about "using a tool too much," sit with that for a second.
Heavy, honest use looks like waste only until you see that cache line.
Most of what looked expensive was work I had already paid for once, reused.
This is the part people get wrong when they see a big usage number and reach for the word "abuse."
Nobody games a subscription for two months straight.
You show up every day and let the tool carry the parts a person should not have to carry by hand.
So why did I still feel behind
Because output kept masquerading as progress, and I fell for it my whole life.
I could point at 20.7 billion tokens of proof that I did the work.
And I still froze at the one thing that actually moves a career forward.
Publishing.
Telling anyone.
Standing in the open and saying "here is what I made."
The building was never the scary part.
The building is private, unjudged, safe.
The scary part is the publish button on the post you are reading right now.
I have shipped things quietly for years and let the loudest, most confident, least careful people take the room.
The receipts did not fix that.
No number fixes that.
What actually changed
I stopped treating output as the finish line.
A finished thing that nobody sees stays a private hobby wearing a work costume.
So the new rule is small and it is brutal.
One exposed thing a day.
Not one more feature.
Not one more billion tokens.
One thing that a stranger can see and judge, sent while it still feels too early.
This post is today's.
The building was easy.
This sentence is the hard part.
Your turn
What is the one thing you have built but never shown anyone?
If this was useful
I work through this in public, the wins and the freezes both, mostly on LinkedIn and YouTube. If the real version of building in the open is useful to you, that is where it lives. Find me on X, GitHub, and the work at next8n.com.
Top comments (0)