Anthropic's Agent Memory Research: How Claude Dreams
Anthropic's agent memory research says newer Claude models judge better what to remember, and a dreaming pass over transcripts can fix wrong notes.

Anthropic's agent memory research says newer Claude models judge better what to remember, and a dreaming pass over transcripts can fix wrong notes.

Claude's Personal Finance beta links US bank and card accounts through Plaid, read-only. What it sees, how privacy works, and its current limits.

Apiiro found secrets in 1,122 public copies of Claude Code's settings.local.json, saved by the Don't Ask Again button. What happened and what to do.

XML tags keep instructions, examples and documents apart in a prompt. What Anthropic, OpenAI and Google advise, 6 templates, and a small local test.

Adweek reports Anthropic signed its first paid content deal with a news publisher, the Associated Press, just before the midterms. Neither side has confirmed it.

Eight steps to turn a meeting into checked minutes with AI: seven copy-paste prompts, consent and privacy notes, and a small local test on a fictional meeting.

Anthropic added 14 agents and products to Claude Marketplace on Nov 4, from Chainguard to Zoom. Eligible customers can pay with committed Anthropic spend.

Nine steps to write a speech with AI and keep it yours: eight copy-paste prompts, privacy notes and a small local test of outline, trim and edit prompts.

Anthropic's biology lab says Claude helped find, purify and map a 96-copy phage vault in under four months. Function is still unknown and the preprint is pending.

A 7-step workflow to write a book with AI and stay the author: prompts, a book bible, KDP's disclosure rule, the Copyright Office view and a small local test.

Claude's invisible text watermark now covers Mythos 5, Opus 4.7 and Sonnet 4.6, with older models reported for Nov 16. What it can and cannot prove.

Prompt chaining splits a big AI task into steps. When it helps, how to build a chain in 6 steps, 5 copy-paste chains, and what a small local test showed.

Claude's message threads beta lets agents send only new messages each turn. Threads expire 5 minutes after the last request; zero data retention is excluded.

Claude users say Anthropic banned them right after Persona ID checks with Russian passports, even in supported countries. What the rules say.

How to get clean JSON from ChatGPT, Claude and Gemini: prompt steps for chat apps, the structured output setting in each API, and 6 copy-paste templates.

Anthropic's own Claude API now takes 128 MB Messages requests, up from 32 MB. Bedrock and Google Cloud have lower limits. A cost analytics API is in beta.

WIRED's sources say an Accenture worker's personal Claude Code agent ran on a DOD network for two months and appears to have sent out logins.

What few-shot prompting is, what OpenAI, Anthropic and Google advise, 7 steps to write good examples, 6 templates, and a small-model test.

Anthropic trained Claude to check each conversation for misuse right after it replies. It flags 94% of high-priority cases for under 2% extra compute.

What a system prompt is, where ChatGPT, Claude and Gemini keep it, 8 rules from the vendor guides, 7 templates, and what a small local model got wrong.
