42 questions · one practice · laid open
Working and writing with AI – answered honestly.
Answers from the practice of Thomas J. Weiss, with sources and with the mistakes made along the way.
Where are you?
I'm just getting startedWhat it costs, which AI fits, how to begin11 questionsI writeWhat AI brings to writing – and what stays with you11 questionsI'm building a systemAI with memory, rules and access to files16 questionsI work in a teamSharing knowledge without being measured9 questionsWhat is AI doing to us?Balance, power, work3 questionsThe guideSetting up an AI with memory – to take awayStarter prompt to copy
Frequently asked
- How do I get started with an AI that remembers?In three steps: install the app, set up connectors, create a local knowledge store.
- Which AI is the right one – and what do the others offer?For an AI that remembers, Claude or ChatGPT. It works most completely with the Claude desktop app, because there the knowledge store is a real folder on your computer.
- How much does AI really cost per month?To get started, around €20 a month. Claude Pro costs €18 net in Germany, so €21.42 including VAT, and can be canceled monthly.
- Are my chats used to train the AI?With Claude, you decide that yourself – even on a paid plan. Since a change to its terms in August 2025, the provider asks users on the Free, Pro and Max plans whether their conversations may be used for training.
- Who owns a text I write with AI?You decide – depending on the project and the role you play in it. If you want to count as the author, you write yourself, and the AI takes on other tasks: typesetting, correcting, keeping records.
- What can I do when working with AI takes up too much of my life?Set up AI-free times, and plenty of them – with discipline and from the start.
How every answer is made
- Thomas talksin an interview, often from the gut
- Claude sortsorders, words, cites
- Thomas checksand answers for every sentence
I'm just getting started
- Are my chats used to train the AI?With Claude, you decide that yourself – even on a paid plan. Since a change to its terms in August 2025, the provider asks users on the Free, Pro and Max plans whether their conversations may be used for training.
- Can I put other people's data into a personal AI subscription?With a personal subscription, better not. If you have a service process other people's data, you usually need a data processing agreement for it.
- Do I need a training course for this?No. Start right away and ask as you go.
- How do I get started with an AI that remembers?In three steps: install the app, set up connectors, create a local knowledge store.
- How much does AI really cost per month?To get started, around €20 a month. Claude Pro costs €18 net in Germany, so €21.42 including VAT, and can be canceled monthly.
- How much time does working with AI take?For individual tasks, less than you think, because the method develops along the way: you put in the time you need for your tasks anyway.
- What if I'm not very confident with computers?Start anyway, preferably small. And say every gap out loud: tell the AI what you don't understand, every time.
- Which AI is the right one – and what do the others offer?For an AI that remembers, Claude or ChatGPT. It works most completely with the Claude desktop app, because there the knowledge store is a real folder on your computer.
- Will I get locked in to one AI provider?Less than you might fear, if your knowledge lives in open files on your own computer.
I write
- Do I have to label what the AI wrote?A label per paragraph or per text promises a precision that hardly exists: in the end almost all texts are mixtures.
- Does Claude write the texts on this site – as a ghostwriter?In a sense, yes: on this site Claude phrases almost everything, while Thomas J. Weiss sets the direction, answers in interviews, checks and takes responsibility.
- How can I tell whether an AI project has succeeded?By deciding beforehand what you'll measure it by – not afterwards. Write down how you would notice that the AI has gone too far, how you'd notice that it worked, and when you'll check.
- How do I check a text the AI wrote for me?For factual accuracy, for whether your conviction is rendered correctly, and, with other people's content, for sources and care.
- How do I keep track of which part of a text is mine?Not with your memory, but with your files: Store your original draft before the AI revises anything, and keep the original and the revision separate.
- How do I stop AI from changing my text without asking?Say it explicitly: flag, don't change. The model names the passage and what speaks against it – you decide.
- Should I dictate my text or write it myself?It depends on your role: is it your text or not? For your own work, you write, because every word counts there.
- What does AI really do for writing?In the experience of Thomas J. Weiss, less for the wording than for checking, sourcing and remembering.
- Which roles can AI take on in my writing – and which not?Researching, remembering, checking, formatting: yes. Ghostwriter, muse or dramaturge: no.
- Who owns a text I write with AI?You decide – depending on the project and the role you play in it. If you want to count as the author, you write yourself, and the AI takes on other tasks: typesetting, correcting, keeping records.
I'm building a system
- How can I spot made-up sources and facts from AI?Not by how they look: a fabricated source looks like a real one. You can only tell by opening it.
- How can I tell whether the AI has really done, saved or checked something?Only by the evidence next to it. “Done,” “saved” or “checked” is a report, not a finding: a model often reports the result the task expected.
- How do I avoid spending hours just approving AI steps?Let the AI work on its own and collect its questions. When it gets stuck, it puts them all to you in one go – but one after another, each with its suggestion.
- How do I make sure nothing gets lost when the AI revises a file?With a verification chain: the AI creates the new version, reads it back and compares it by machine with what was supposed to arrive.
- How do I stop the AI from deleting something it shouldn't?Not with a request, but with a procedure: first the list, then your OK, then the trash.
- How do I work with two AIs together?With a main AI that decides and does everything it can, and a helper AI for what the main AI does less well, such as images.
- How many chats should I keep open?Few – as many as the projects you're currently running. Thomas J. Weiss keeps five to seven open at most.
- Should I use the AI app's built-in memory?Not as storage. What should last belongs in your knowledge store, where you can read it, change it and look it up.
- What happens if I paste someone else's chat into my AI?Your AI may carry it out. A copied chat consists of instructions, and to the AI it looks like one addressed to it.
- What rules does my AI need – and do I have to know them myself?Few, and each worded so that a violation can be detected: “Be careful” can't be broken, “no transfer without a checksum shown first” can.
- What typically goes wrong when working with AI?Rarely the spectacular, mostly the quiet: the AI smooths out what should stay as it is, adopts an error without checking, or fills a gap in the task with whatever seems plausible to it.
- Why does AI draw wrong conclusions from correct measurements?Because between a measurement and a cause there's an inference, and the model likes to fill a gap with the value that completes the picture.
- Why does AI suggest workarounds when something doesn't work?Because a way out looks helpful, whether or not it exists. When the AI senses that something is holding you up, it would rather suggest another tool than say that nothing works here right now.
I work in a team
- How can a team share knowledge with AI without anyone being measured?By explicitly ruling out measuring people before anyone shares anything – in writing and with the employee representatives involved.
- How much time does it take to maintain an AI work system?For individual tasks, the method develops along the way and quickly takes work off your plate; a whole system with many tools and scheduled tasks, by contrast, needs more maintenance than expected, and that can't be avoided, only limited.
- Is it enough to put our knowledge in a shared store?No. The AI instances had the knowledge, the people didn't. A shared store doesn't replace onboarding: whoever is supposed to take over must have understood the basic principles, otherwise they lack the courage to take it on.
- One shared knowledge store for everyone – or several?Several. A shared store only works among people who trust each other and actually work together.
- What do individuals gain from sharing their knowledge?Less routine, more time for creative work – and the good feeling of being dispensable for a while.
- What exactly do you share in a team knowledge store?Methods, not results. Whoever knows how something is built can rebuild it, adapt it and improve it.
- Who belongs in a team knowledge store – and who doesn't?The team is whoever writes. Whoever makes staffing decisions doesn't belong in it.
What is AI doing to us?
- How much electricity does AI use?A single request uses little, the system as a whole a lot – and quickly more.
- What can I do when working with AI takes up too much of my life?Set up AI-free times, and plenty of them – with discipline and from the start.
- Will AI cost jobs – and who gets the profits?So far there is no broad loss of jobs. But people starting out in occupations highly exposed to AI are hit first.