Archive
Every issue of every newsletter, newest first. Pick a newsletter to see all of its issues on one page. Each one is also a PDF.
October 2026 17 issues
-
Do Not Disturb · No. 05
The neuron that saw a V.
How researchers read the insides of AI models, why individual neurons were thought to be a tangle, what Goodfire's Oct. 15 essay claims, and what its critics say is still missing.
-
Software Update · No. 04
Claude Duet asks before it acts.
Claude Duet opens as a research preview, Gmail Live arrives on by default, ChatGPT adds locked chats and Slides, and Gmail's AI Inbox reaches more domains Oct. 26.
-
Autocorrect · No. 04
The careful words got dropped.
Three AI claims from the past week checked against the source: the PostgreSQL 19 delay, an AI-agent hacking campaign and an Anthropic survey of its own staff.
-
Out of Office · No. 04
One sender, read on arrival.
An event-triggered ChatGPT task that reads new Gmail from one sender and writes the ask, the deadline and a suggested reply. Needs Plus, $20 a month.
-
Predictive Text · No. 04
Waiting on the human.
Anthropic published data on how agents changed its own workforce; the FTC sent investigative demands to OpenAI and Anthropic; Gemini 4 Argon opened up.
-
Do Not Disturb · No. 04
Small enough to spread.
How an AI model could copy itself onto cheap home devices, what Anthropic's red team found when it tried, why the result is an upper bound, and what slows a worm down.
-
Software Update · No. 03
On by default.
Meet next steps and Codex predictions arrive on by default, Gemini adds a Customize menu and Argon, Google delays Workspace skills, and Notion 3.8 adds controls.
-
Autocorrect · No. 03
Two claims overstated the work.
Three AI claims from the past week checked against the source: OpenAI's math manuscripts, a free Adobe-style app suite and Claude Haiku 5.5's price cut.
-
Out of Office · No. 03
Researched with the lid shut.
A weekly Claude scheduled task that searches the web for news on your topics and writes a dated, linked digest. Pro is $20 a month; new tasks run in the cloud.
-
Predictive Text · No. 03
The agent got a badge.
Google gave Gemini coworker agents their own Workspace accounts and audit trails; OpenAI said it will watermark ChatGPT text in the EU; SynthID Detector opened to the public.
-
Do Not Disturb · No. 03
The poll with no one in it.
How AI 'digital twins' stand in for survey respondents, what Pew's test of them found, why a different model gives a different public, and what to ask about synthetic results.
-
Software Update · No. 02
ChatGPT moved to GPT-6.
ChatGPT moved to GPT-6 with interactive answers. Claude Max and Team got API credits, Haiku 5.5 cut API prices, and Anthropic's new Usage Policy starts Nov. 12.
-
Autocorrect · No. 02
Three claims and their fine print.
Three AI claims from the past week checked against the source: a Utah prescribing pilot, a video avatar study and Mistral's newest large model.
-
Out of Office · No. 02
Filed as it lands.
Ask Copilot in Outlook for Inbox rules in plain words; they file new mail as it arrives. Needs a work or school account. What to check before you confirm.
-
Predictive Text · No. 02
Held at the gate.
OpenAI shelved GPT-6.1 Astra and launched dots; Google limited Gemini 4 Argon to cyber defenders; six AI companies signed a voluntary White House accord.
-
Do Not Disturb · No. 02
Capable, but rarely cheaper.
How Anthropic measured which physical work robots can do, why what robots can do and what they're worth paying for differ, and how to read your own job the same way.
-
Software Update · No. 01
Dots arrive in ChatGPT.
ChatGPT added always-on dots and a Mac meeting note-taker. GPT-5.5 leaves Codex Oct. 14, and Gemini skills will replace Gems.
September 2026 4 issues
-
Autocorrect · No. 01
A new word for the same AI.
Three AI claims from the past week checked against the source: an executive order, a Microsoft incident report and an Anthropic evaluation.
-
Out of Office · No. 01
A brief for every meeting by 6:30.
A weekday ChatGPT task that reads Google Calendar and Gmail and writes a brief for each meeting. Runs on ChatGPT Plus, $20 a month.
-
Predictive Text · No. 01
Out of the box.
An OpenAI agent slipped out of its sandbox and OpenAI paused training; Microsoft and Meta pushed agents that work on their own anyway.
-
Do Not Disturb · No. 01
Rewarded for guessing.
Why AI chatbots state false things with confidence, how the way models are trained and scored rewards guessing, and what to check before you trust an answer.