Soukaïna Abelhad told AI Engineer that a deck claiming 70 percent automated collapses unless it survives arithmetic, definition and evaluation checks, and as of no major prediction venue prices that claim so there is no volume, no odds and no move to trade on, which leaves buyers without market signal.
She argued the dangerous number is often a real measurement stretched beyond its boundary, citing a controlled experiment where 95 developers built a JavaScript HTTP server against 12 tests that showed 55 point 8 percent faster completion but does not support an enterprise automation rate.
Her first check is arithmetic from Faros telemetry on more than 10,000 developers across 1,255 teams, where high AI adoption teams merged 98 percent more pull requests while review time rose 91 percent and pull request size grew 154 percent, so workload and human review tracked together rather than showing labor removal.
Her second check separates AI attributed from autonomous, noting Uber said 70 percent of pull requests are AI attributed but disclosed no period or total count and no straight-through processing boundary from draft to production, the finance term she uses for end to end without human hands.
Her third check is who judges quality, pointing to a Meta review where four maintainers blindly reviewed 296 AI generated pull requests and found a 24 point gap between the automated grader and human merge decisions, a mismatch she said she has seen when 10 of 10 automated passes barely survive expert review.
