The line gets read as hype. It is an instruction about expiry dates.

Every opinion you hold about a tool carries a timestamp. You tried the model on a task last year — it mangled the numbers, invented a citation, missed the point — and you filed a verdict: not good enough for this. The verdict was correct. It was also perishable, and unlike the milk, it never announced when it went off.

Most tools sit still while you learn them. The spreadsheet you mastered a decade ago is, for the most part, the spreadsheet today; your competence compounds quietly on stable ground. This technology does the opposite. It changes under your feet every few months, which means your hard-won map of what it cannot do is the fastest-depreciating asset you own. The person still deciding by their 2023 test is not being careful. They are navigating with an old chart.

Notice where this bites. It is not the tasks you delegate — those get re-checked by use. It is the tasks you ruled out. The analysis you still assemble by hand because it "can't handle nuance." The draft you never hand over because it "doesn't sound like you." Each rule was a finding once. Now it is a habit wearing a finding's clothes.

Mollick's sentence turns the ratchet the other way: whatever fails today is the floor, not the ceiling. Plan on the floor rising.

So today: pick one task you decided the machine could not do — one specific job, ruled out months ago — and run the test again. Twenty minutes. If it still fails, your verdict earned another quarter. If it passes, you just found out how long you had been working for a rule that expired.