the meta AI number that stuck with me isn't a headcount. it's this: code changes to meta's internal platforms and infrastructure up 220% year over year, while changes that actually shipped as user-facing features were up 36%. that's from an internal post by meta's CTO, reported by reuters. so a lot more output, not a lot more product. somebody still has to review, fix and clean up all that extra code, and that cost almost never shows up in the AI budget. it fits what reuters reported back on aug 26. under "project OT", meta explored cutting some teams by up to 60% (meta says that was never 60% of the whole company), laid off ~10% in may, then cancelled the second wave the day before it was due. it has since dropped AI adoption from performance reviews, per the information. microsoft's kathleen hogan said it plainly: "We learned that access and usage do not equal transformation." none of this means meta is walking away from AI. it's still spending heavily on it, so this looks more like a timeline reset than a retreat. but if i were running an AI rollout right now, i'd stop tracking how much people use the tools and start tracking what actually ships.