May 18, 2026 | Issue 51
One subtraction ➖ One analogy 🥛
Created by Sam Rogers, building PAICE.work | Freely available on Substack and LinkedIn | New issue every Monday
🔭 Signal: Permanent = Expired
There is a date stamped on the milk when it’s packaged. You don’t open the jug to check it. You tilt it, read the stamp, and toss as needed. You never had to validate the chunks.
In 2018 I wrote an article with this issue’s title. It was about training assets, the documents and courses we build as if they will be permanent when we know damn well they won’t be. Eight years on, the line I stand by hardest: there is a 100% chance they will not stay put. The only open question is how long the rot takes.
I was aiming too small back then. It was never just the training catalog. All knowledge is an assertion with a shelf life. A fact is less a fixed object than a snapshot of where our understanding happened to be standing the moment we looked.
What does your favorite AI tool actually remember about you? Nothing. It rebuilt everything it appeared to know from the context in front of it, in real time, and next time it will do that again from zero. The weights were baked in at birth and never move. The most knowledgeable system ever built does not learn and does not store. It pours fresh every single time. That is not the flaw. That is the design that keeps it current while your saved playbook quietly turns in the dark.
Case in point: in the last two weeks the two most cited AI regulations on the planet moved, and not by inches. Every brief, policy, and rollout plan written to the old text was correct in April and is wrong in May. Same words on the page. The date underneath them rotted away. Watch the video here.
➖ Subtraction: No need to smell what stinks
Here’s the power move, and the discipline is in what it forbids.
Find one knowledge artifact you still serve from whose date is just plain old. The onboarding deck from two reorgs ago. The prompt set from before the last model. The wiki page nobody has edited in a year. Anything that mentions COVID-19. The framework in your head.
Kill it this week without opening it. This will feel reckless. Do it anyway.
Do not audit it first. Do not build the case for why it failed. Do not run it one more time to be sure. The age IS the verdict. Demanding proof of spoilage before you’ll throw anything out is exactly how the graveyard filled up in the first place.
The harsh truth is that you have not been maintaining a system. You have been serving spoiled stock under good lighting and calling it institutional knowledge. And the math has flipped: re-examining the old carton now costs more than pouring a fresh one.
So read the date. Toss what expired. Move on.
🥛Analogy of the Week: Milk Jug in the Fridge Door
The milk jug does not announce anything. No blinking light, no dashboard turning red. It just turns, silently, on a schedule it set the day it was filled and impacted by what changed around it ever since.
You handle this perfectly in your own kitchen. No tasting panel. No quota of chunks required before you’re allowed to pour it out. You read the stamp and you decide. The date can’t prove the milk is still good. It can absolutely prove it’s done.
Every knowledge store you keep works exactly like this jug. The 2021 training doc is a few years sour. Your AI mental model now turns in weeks, and the interval shortens every quarter.
You don’t have to smell last month’s milk to know it’s bad. You only smell this month’s to know if it’s still good.
🎵 Closing Notes
If everything spoils, what are you even supposed to hold onto?
One thing. The only input with no best-by date is the disposition to keep pouring fresh. Curiosity does not expire, because it was never stored. It is not a stock you defend. It is the act that regenerates the knowing on demand, without grief over what you let go.
So the question that should sit with you this week is not what to learn next. It is this: what are you still serving from, long past its date, because throwing it out would mean admitting it was never permanent?
You already know how to read a date. You do it at the fridge every day without a second thought. The hard part was never the content.
Until next Monday,
Sam Rogers Curiosity Stocker
P.S. Thanks to Keith, Evan, and everyone else who gave feedback after last week’s invitation. Next week marks one year of this weekly newsletter, and I’d still love to hear from you! Reply with a quick sentence on what you’d like to see next. What would you find most valuable?
Need to know which AI regulations just moved, which cases are still pending, and want it in a form a machine can actually read on demand? That is the PAICE Portfolio’s Legal Graph layer. It pours the current answer fresh, so your policy isn’t the carton in the back of the fridge.