To answer an RFP, tender or security questionnaire with AI, give it the questionnaire as the structure and your current policies as the only source, require a quote from a named document under every answer, and check every answer that states a number, date, certification or promise. Treat any answer without a quote as a question for a colleague, not a draft to send.

Drafting is the easy part now. The expensive mistakes are elsewhere.

SCAMPER is a checklist of seven prompts for improving an existing idea, product or process: Substitute, Combine, Adapt, Modify, Put to another use, Eliminate, Reverse. You run the idea through each prompt in turn and write down what it suggests. It works because it forces you past the two or three moves you would make anyway.

Part of Thinking models in practice, a short series on classic methods and how to run them on your own material.

To turn voice notes into Markdown without breaking your flow, use a global shortcut that opens a capture box over whatever app you are in, read the transcript before it saves, and save it to today's note as plain text. Only then — and only if you want — let AI tidy how you said it, never what you said.

The usual voice-memo workflow fails at the second step. You record, the audio sits in an app you never open, and the transcript, when you finally get one, is a wall of "um, so, the thing is".

First principles thinking means breaking a problem down to the things you are confident are true — physical limits, real costs, actual requirements — and reasoning up from those, instead of starting from how the problem is usually solved. It is slow and effortful, so it is worth using only where convention may be wrong and the stakes justify the work.

Part of Thinking models in practice, a short series on classic methods and how to run them on your own material.

An AI skill is a plain-text file — usually SKILL.md — with a name and a one-line description of when it applies at the top, and the method below: the steps, the checks, the shape of the output. To write one that gets used, describe the situation it applies to rather than its contents, keep the method concrete enough to be followed literally, and fix it the first time it lets you down.

The third time you type the same instruction into a chat box — "decisions first, each with an owner and a date" — that sentence has become a skill. It just has not been saved yet.

To study for an exam from lecture slides with AI, first rebuild what the lecturer said around each slide, then ask about the exact diagram or formula you did not follow, then switch from reading to being tested — answering questions before seeing answers. Reading AI explanations feels like learning; being quizzed is what tells you whether it was.

Slides are a strange thing to revise from. They are the lecturer's prompts, not the lecture. The half that explained them lived in someone's head and is gone.

To steelman an argument is to restate the opposing view in its strongest form — better than its own advocates might — before you respond to it. It is the opposite of a straw man. If your position still stands against the steelman, it is worth holding; if it does not, you found out cheaply.

Part of Thinking models in practice, a short series on classic methods and how to run them on your own material.

AI edits to your writing should arrive as tracked changes — deletions struck through, insertions marked, inside your own text — so anything without a mark is exactly what you wrote and you can accept or reject each change. The alternative, a rewritten copy, forces a choice between rereading everything and trusting an edit you never saw.

On a paragraph, a rewritten copy is a small annoyance. On a chapter, it is the reason people stop letting AI touch anything that matters.

To read a Kindle book on your desktop with AI, you need a DRM-free .mobi or .azw3 file and a reader that converts it into a proper book with chapters. Open it in MindMax and it becomes an EPUB-style book the first time, with the original file left untouched; from there you can ask questions with quotes, or have an AI annotate it chapter by chapter.

The catch comes first, because it decides whether the rest applies to you: books bought from the Kindle store carry DRM, and they cannot be opened. That lock is not ours to remove, and this post does not explain how to remove it.

A local model through Ollama or LM Studio handles short, contained document work well — explaining a passage, answering a question about an open paper, tidying a note — and that work never leaves your machine. It struggles with long jobs that need a large context and reliable tool use: reading a whole book chapter by chapter, translating hundreds of pages, drafting something book-length.

That split is worth knowing before you start, rather than four hours into a job that will not finish.