"You can't legislate your way out of a bad incentive. You can only out-design it — fast." — Aaron Bare
Markets clear in milliseconds. Models retrain overnight. Memes mutate faster than policy. The substrate underneath every leadership decision is no longer stability — it's acceleration. If your operating system was tuned for 1995, it's currently on fire.
A small misalignment at human pace is an annoyance. The same misalignment at machine pace is a market. What used to take a decade to rot — call it slow-cooked Enron — now ferments in a quarter. Compounding is neutral. The thing being compounded is not.
Engagement bait was a content problem. At AI scale, it's a civilizational one. Quarterly thinking was a finance quirk. At AI scale, it gets baked into the reward function of the systems writing your strategy. Speed promotes every old sin to a new category.
Regulation is a downstream lever. By the time the policy ships, the incentive has already shipped ten cycles of behavior. The only durable move is upstream — change what the system rewards. Govern the loop, not the output.
Every cycle you don't root out an incentive, it metastasizes into the next layer of the stack: from product to platform, from platform to protocol, from protocol to norm. The hardest thing to root out is something that has become invisible because it became universal.
"Slow systems forgive bad incentives. Fast systems don't. That's the whole job now."
— From It's Really Not That Serious
Drag the dial. As things speed up, hidden incentives metastasize. Defuse each one with a root-out move. Clear the board to see what a clean system looks like.
Make the incentive visible. Write it on the whiteboard. If you can't name it in one sentence, you don't yet see it.
Wells Fargo's cross-sell quota lived in a sales-ops doc for years. Naming it ended a $3B fraud.
Follow the reward to the behavior it actually produces — not the behavior it was designed for. Behavior is the receipt.
The Hanoi rat bounty produced rat farms. The reward was for tails, not for fewer rats.
Simulate the incentive at 10x volume. If it breaks the system at scale, it's already breaking it at scale — you just haven't noticed.
RLHF's sycophancy bias was a quirk at training scale and a worldview at deployment scale.
Tie the reward to the outcome you actually want — not the proxy that's easy to measure. Cheap proxies create expensive behavior.
YouTube re-anchored from raw watch-time toward 'satisfied watch-time' once the cost of attention-maxxing came due.
Speed demands cadence. Schedule the next audit before you finish the current one. Incentives drift; calendars don't.
Boeing's MCAS reviews fell off a cadence. The plane didn't.
The 6 labs catalog the patterns. The Root-Out Framework is what you do with the catalog. Start where you suspect the rot is loudest.