The memos
Detailed, sourced guides on AI agents and the tools around them. Free to read, written to be used.
Build the Claude Code skill that makes your videos
The reel that sent you here was made by Claude Code running one skill: a plain text file. Four steps to write your own for any job you repeat: the phases, the three stops, the tools with their exact commands, and the mistakes section that makes it good. My pipeline is the worked example, with a template to fill in.
Read the memo →PICKERClaude Sonnet vs Opus vs Fable: which to pick
The instinct is to tap the biggest model and leave it there. Anthropic's own guide says start on Sonnet. Which model is for which job, why every step up eats your limit faster, what each plan really includes (on Pro, Fable bills extra from the first message), and the four topics Fable hands down to Opus.
Read the memo →SCANScan any AI skill before you install it
A quarter of the skills people install into their AI carry a security hole, and one in twenty looks deliberately malicious. NVIDIA released a free scanner. The command, how to read the 0-100 score, the five things it cannot see, and the four-second rule that needs no tool at all.
Read the memo →CLASSROOMClaude for Teachers: now free for whole schools
Anthropic just made the paid Claude free for entire US K-12 schools and districts. An administrator verifies the school and every teacher gets it. What it actually does on a Sunday night, which of the two sign-up doors is yours, and the deadline that ends the free year.
Read the memo →CHROMEWhat is Claude in Chrome? Everything it does
Claude can now read, type, click and fill in forms on any site, using the logins you are already signed in with. The real change is that it stopped asking permission first. What it does, the three install steps, and the five places I would not point it.
Read the memo →TOPICSWhat Claude remembers about you, and how to wipe it
Claude's memory is not a black box. It is a list of files you can open, read and edit, and as of 25 August it writes them while you talk. Where the list lives, the five things to check first, and why deleting the chat leaves the note behind.
Read the memo →MONDAYThe nine rules behind Anthropic's Monday send
An Anthropic marketer swapped Sunday-night slide decks for a Claude Code job that messages every sales rep on Monday. The automation is not the asset: the nine rules he added when it got things wrong are. All nine, plus a one-spreadsheet version.
Read the memo →ACADEMYClaude Academy and the 4D AI fluency framework
Anthropic put the AI training it gives every new hire online, free and with no account. The four habits it teaches in plain English, the line where they admit prompt tricks are expiring, what to take first, and what to skip.
Read the memo →LOCALMake AI videos on your own computer, for free
The same kind of model runs on your own machine, offline, with no subscription. The hardware that actually works on both platforms, the one prompt that installs the lot, the settings for your memory, and honest generation times.
Read the memo →TEENSWhat is ChatGPT for Teens? Everything that changes
OpenAI split ChatGPT in two, and under-18 accounts are moved across automatically. The four signals that decide, what gets restricted, all ten parental settings with their defaults, and the way out if it has your age wrong.
Read the memo →LABClaude Science explained: how it earns trust
Anthropic built an AI for scientists, and 79% of them named trust as the barrier. The five design choices that answer that, the 60+ databases, the three-phase rollout, and the fine print the FAQ buries.
Read the memo →FRAMESCan Claude watch videos? The /watch skill, set up
It downloads the video, cuts it into still frames and reads them with the transcript. Nothing is ever played. Both install paths, all four detail levels, and the gaps that decide what Claude can actually see.
Read the memo →TOKENSClaude Code token usage: six habits that cut it
Claude Code re-sends your whole conversation on every message. The six habits that keep it small, the exact commands, how long the cache actually lasts, and everything that quietly resets it.
Read the memo →MAPWhere Google's AI can't reach
AI Overviews halve your clicks, but not evenly. The measured trigger rate for every query type, why a name search is the deepest blind spot at 0.9%, and how to be the source the AI quotes.
Read the memo →WATERMARKThe Claude watermark remover, and what it actually strips
Claude watermarks its text now, and the free removers delete invisible characters. The watermark is not a character. What genuinely comes off, and the one tool worth installing.
Read the memo →CLONEClone any website with one command: the whole setup
A free template and one line rebuild any page as a real project you own. The exact steps, plus the four things a browser never receives and therefore nothing can copy.
Read the memo →SNAKEVibe coding for complete beginners: build a real game from zero
Install the two tools, make a folder, and describe what you want. A working program you own in fifteen minutes, and the point where not reading the code stops being fine.
Read the memo →MEMORYWhat ChatGPT actually knows about you (and how to delete it)
The viral “secretly knows about you” trick works, but it is not a secret. The page it reads, and the five sources you have to clear to actually delete anything.
Read the memo →LOOPStop prompting. Build your first AI loop.
The creator of Claude Code stopped prompting and started running loops. What a loop really is, and how to build one in twenty minutes.
Read the memo →LINESThere are no secret ChatGPT codes. Here's what actually works.
The viral codes are folklore. These are the one-line instructions that genuinely change answers, and why each one works.
Read the memo →AGENTSend an AI shopping: agent mode, explained honestly
The discount-code trick that went viral, what agent mode really is, and the safety rules the reels skip.
Read the memo →CONFIGThe free file that makes your AI code like a pro (ECC, explained)
An Anthropic contest winner spent 10 months teaching his AI how to work, then gave it away. What ECC does, and the real link.
Read the memo →TEAMStop using one AI. Build your first AI team in an afternoon.
One expensive AI as the boss, cheap AIs as the workers. The setup the best teams use.
Read the memo →SECUREAI built your app in a day. Hackers can break it in two.
The five repairs that take one afternoon, with copy-paste prompts and the test checklist.
Read the memo →