Amazon Just Killed Scorecard AI

Amazon didn't pull its AI leaderboard. It pulled the rip cord on the entire metric-driven adoption playbook, and most of the Fortune 500 is still using the same parachute.

The clean read of the Financial Times reporting is cost discipline. Senior VP Dave Treadwell tells staff "don't use AI just for the sake of using AI," internal costs rose faster than productivity, the leaderboard goes away. Standard operational tightening. Amazon runs lean. This is what lean companies do. Move on.

That read misses what an AI usage leaderboard actually does inside an engineering org. It tells every engineer their performance is being measured by whether they typed into a Copilot window today. Goodhart's Law isn't a clever framing here — it's the literal mechanism. The metric became the target. The target destroyed the signal. The signal was supposed to be productivity, and the productivity went the wrong way hard enough that an SVP wrote a memo. That's not cost discipline. That's an admission that scorecard AI doesn't work.

Now look at what Marc Benioff said the same week. Sales is the one department Salesforce is still hiring for. Reassuring, if you're in sales. Except Salesforce's flagship Agentforce product is explicitly built to automate the workflows sales development reps run today. The hiring signal isn't a permanent exemption. It's a transition lag, dressed up as good news, delivered by a CEO who has every commercial reason to keep sales organizations from canceling his platform out of self-preservation. I don't know whether the hiring claim holds past the next earnings call, and the gap between Benioff's framing and his own product roadmap is where I'd watch. Same correction as Amazon. Opposite spin direction. Identical underlying dynamic: deploy AI, watch the scorecard break, communicate around it.

The New York Times offered Schneider Electric this week as the antidote — French manufacturer, augmentation not replacement, here's how it's going. The reporting is careful. The frame around it isn't. Capital-intensive manufacturing in France didn't have the labor-arbitrage option to begin with. Augmentation was the rational play because replacement wasn't available. Citing Schneider to a Big Four partner or a regional bank executive is reading the conclusion without the conditions. The structural calculus in professional services runs the opposite direction.

What we're watching across all three stories is scorecard AI — adoption mandated, measured, and incentivized by a metric that doesn't capture whether the tool produced value. Amazon ran the experiment and got the cleanest negative result on record. Benioff is running a softer version: a hiring signal positioned as a metric to suggest his product won't hurt the people he sells it to. Both are corrections. Neither is being labeled as one.

Here's the dare. By the end of Q3 2026, three more named Fortune 100 companies will quietly retire an AI adoption metric. None of them will call it a reversal. Watch for the press releases that use "recalibrate," "evolve," or "mature our approach." That's the correction phase for scorecard AI. It's already underway. Amazon just said the quiet part.