Past a point, strategy is just expensive forgetting.

Your smartest people might be your biggest source of drag. Not because they are wrong. Because they spend too many cycles being smart, and every extra cycle lowers the odds you are actually serving a real need.

There is a place for strategy. Past a point, it is just expensive forgetting.

We just saw this simulated with the Opus 4.8 release. The cleanest proof I have seen this year came from an AI lab, not a boardroom. Anthropic's new model beat the old one across the benchmarks. Then they gave both versions $500 and one year to run a real business.

The older, dumber model turned $500 into $10,800. The newer, smarter one turned the same $500 into $3,000.

Here is why. They ran it at maximum reasoning. It spent so many cycles thinking that it filled its own memory and forgot what it had already done. Too busy being smart to stay on task. When they turned the thinking down, it nearly doubled its money. Less strategy. More results.

Your team is not a language model. But the failure mode is identical. Intelligence is not the goal. Serving the need is the goal. Process before prompts.

The $7,800 Gap

Look at the setup again. Same $500 starting capital, same year, same product to sell. The only variable was how hard each model was allowed to think before it acted.

The smarter model ran at maximum reasoning and paid for it. It spent so many tokens re-deriving its own strategy that it pushed earlier decisions out of its working memory. It re-ordered inventory it already had.

It re-litigated pricing calls it had already made three turns earlier.

The older model skipped most of that. It executed, checked the result, and moved on. The gap between the two — $10,800 versus $3,000 — is $7,800 that strategy quietly ate.

The measure of an agent is not how well it reasons. It is how much of what it already decided survives contact with tomorrow.

Where This Shows Up At $12M in Revenue, Not $500

Swap the vending machine for a six-person operations team. The forgetting looks the same, just slower and more expensive. A pricing exception gets approved on Monday, then re-argued from scratch on Thursday because nobody wrote down why.

This is not a knowledge problem. Every person in that meeting is smart enough to remember. It is a process problem — there is no system forcing the decision to stick.

Cloon's discovery work exists to find exactly these loops before anyone talks about agents. In one two-week engagement, mapping where decisions kept getting relitigated surfaced $388K in identified ROI — not from installing a smarter model, but from writing the decision down once and building a workflow that respected it.

The Fix Is Not a Smarter Model

Buying more reasoning does not fix a team that keeps forgetting what it already knows. A workflow that remembers does. That is the whole argument for process before prompts: redesign how the decision gets made and kept, then decide where an agent earns its place inside it.

Where is being smart quietly slowing your team down?