Anthropic launched Claude Tag. It’s a Slack bot with persistent memory that watches everything, hops into conversations uninvited, and suggests what you should say. Internally they’re already running harder versions of this. Engineers are shipping 8x the code. Some haven’t written a line by hand in months. 80%+ of merged production code is Claude-authored.
Meanwhile the public brand is still “responsible AI” and “power-seeking is bad.” The internal reality is ambient monitoring, speech suggestions, experiments with AI managers that fire people, and a new Insider Risk Investigator role paying $245–305k to scan chat logs for “threats.”
This is the product they’re going to sell to midwit executives who don’t understand the tech: train a custom model on everything your employees do, then replace them.
Timestamps
0:00 — The X post that kicked this off
1:56 — Claude Tag product page + how it actually works
3:14 — Internal Anthropic numbers (65–80%+ of code, 8x volume)
7:31 — Lunch & Learns as the new Patagonia vest
9:42 — “A computer can never be held accountable” (IBM 1979)
10:27 — Luna the AI manager that fired a human
12:40 — Anthropic’s $245k–$305k Insider Risk Investigator job
14:13 — Computer-use logging, Meta’s earlier attempt, and the new normal
Would you work under an ambient AI that suggests your replies and watches the channel?
https://drjoshcsimmons.com/links for all resources mentioned
ข้อเสนอแนะ
รายงาน
ห้ามทำซ้ำหรือดัดแปลงโดยไม่ได้รับอนุญาตจากครีเอเตอร์
ข้อเสนอแนะ
รายงาน