Skip to content
Multigrid

AI for people who do not write code

Methods for using AI well without programming: how to ask, how to check the answer, and where the honest limits are.

Most guidance written for non-technical readers is a list of products. Lists of products go out of date faster than they can be written, and they teach nothing that transfers: a person who has memorised where the buttons are in one tool is helpless in the next one, and a person who understands what the thing underneath is doing is not.

These pages teach methods. Each one covers a task an ordinary working person actually has — write a difficult email, read a hundred-page report, prepare for a meeting, check whether an answer is true — and gives a procedure, a worked example, and a way to tell whether the result is any good. Products are named only where a sentence would be meaningless without the name.

Two things run through the whole cluster. The first is verification: every method here ends with a step that tells you whether it worked, because a plausible wrong answer is the characteristic failure of this technology and it does not announce itself. The second is that some tasks should not be handed over at all, which is a separate page and also a sentence in most of the others.

How to Write a Prompt That Gets a Useful Answer

Five moves that turn a vague request into one that gets a usable answer, each shown as a before-and-after on the same real task.

9 min read

Getting an AI to Work With Your Own Documents

What actually happens when you upload a file, why it sometimes misses text that is definitely in there, and the questions to answer before you send anything.

10 min read

Summarising a Long Report Without Losing the Point

A two-pass method that extracts figures and caveats before it writes anything, so the summary keeps the parts a one-shot summary always drops.

9 min read

Using AI to Prepare for a Meeting

Turning the material you already have into a briefing, a question list and a set of likely objections, with a source attached to every line.

9 min read

Fact-Checking What an AI Told You

A five-minute routine that triages which claims need checking, and the specific checks that catch the errors this technology actually makes.

9 min read

Spotting AI-Written Text Without a Detector

Which textual signals are weak evidence, which are worthless, and why neither a detector nor a careful reading may be used to accuse anybody.

9 min read

Using AI for Job Applications, Honestly

Where the line is between help and misrepresentation, what actually improves an application, and what screeners and interviewers catch.

9 min read

AI for Studying: Methods That Aid Learning

Five study methods built on retrieval practice, where the model tests you rather than answering for you, with the prompts and the failure to watch for.

9 min read

Designing Assignments in a World With AI

Four properties that make a task resistant to being outsourced, three assignments redesigned end to end, and the rule you have to write down.

10 min read

AI for Small Business Admin

A test for which paperwork is safe to hand over, a method for five common tasks, and the arithmetic to work out whether it is worth it for you.

10 min read

Writing Emails With AI That Still Sound Like You

How to build a voice sample from your own sent folder, turn it into rules a model can follow, and keep the whole thing private.

9 min read

Making Sense of a Spreadsheet Without Knowing SQL

How to get from a question to an answer you can trust, including the verification step that separates a real number from a plausible one.

10 min read

Translating and Localising Responsibly

Three tiers of translation need, the checks that make machine output safe at each, and the categories where a qualified human must sign it off.

9 min read

Accessibility Work AI Is Genuinely Good At

Alt text, captions and plain-language rewriting, each with the prompt that produces something usable and the check that says whether it is.

9 min read

AI for Personal Finance Admin, Not Advice

Categorising and summarising your own records is admin and works well. Recommending what to do with your money is regulated advice, and this page stops there.

9 min read

Preparing for a Difficult Conversation

How to set up a rehearsal that is actually adversarial, get the counter-arguments you have not thought of, and avoid the confidence a simulation manufactures.

9 min read

Using AI Without Handing Over Your Data

Why not training on your data and not keeping your data are different promises, the four questions that work on any product, and what no setting can undo.

9 min read

What Free AI Tiers Really Allow

The five limits every free tier is assembled from, where in a vendor's own documents to read the current values, and what you are exchanging for it.

9 min read

Teaching a Team to Use AI Well in Two Hours

A two-hour session you can run without being an expert: a timed outline, three exercises on the team's own work, and a house rulebook they write themselves.

10 min read

The Tasks Where AI Is the Wrong Choice

A short list of tasks where the downside of being wrong is far larger than the time saved, with the reasoning that generalises to cases not on the list.

9 min read

Other topics