Mastering the machine, protecting the mind

Two truths about AI

AI is the most useful tool professionals have been handed in a generation. Used well, it kills the blank page, speeds up the low-value parts of your work, and gives back time you didn't know you were losing.

Used without thinking, it can quietly erode your edge, your credibility and even your free time. And if we're careless, it can make it harder for our children to stay with a hard problem until they crack it. This guide is for people who want the first without the second.

The one rule behind all the others

Treat your AI assistant as scaffolding, not a substitute. Scaffolding holds a building up while it's built, then comes down once the building stands on its own. Used as support, AI makes you better. Used as a replacement, it makes you duller.

Once you count the time spent fixing confident mistakes, the saving on complex work is often smaller than people expect. A 2025 MIT Media Lab study (Kosmyna et al., a preprint of 54 participants) found that people who wrote essays with an AI chatbot showed the weakest brain connectivity of the groups tested, and struggled to accurately quote their own essays, compared with people who wrote unaided. AI is powerful for people who stay in charge, and risky for people who don't.

Sources: Kosmyna et al., MIT Media Lab (arXiv, 2025)

The settings worth two minutes of your time

Menu names change often, so treat these as a guide and check your own app. Most live under Settings.

• Memory (Settings → Personalisation): your assistant keeps saved facts about you. Read what's saved today and delete anything you wouldn't want a colleague to see.
• Custom Instructions: fill it in once — what the assistant should know about you and how it should reply. It shapes every new chat and is the most valuable setting.
• Personality and tone: pick a response style that pushes back on you, not one that just agrees.
• Projects: put recurring work (a client, a course, a report series) in a Project with its own instructions and files, so one client's details don't drift into another chat.
• Temporary Chat: skips memory and chat history. Use it by default for client names, unpublished numbers, health, legal and HR matters.
• Data Controls — "Improve the model for everyone": on personal accounts this is usually on by default, so your chats can train future models. Check it today if you use a personal account for work.
• Export and delete: you can download your data and delete chats or your account. Deleted data may stay on servers for a short time, so make clean-ups a habit.

A custom-instructions starter you can paste

About me: I'm a [role] at a [type of company] in [country]. I mostly use you for [emails / reports / analysis / planning]. My audience is [who reads my work].

How to reply — adapt these lines:

• Be concise. Lead with the answer, then the reasoning.
• Plain English, short paragraphs, no hype or filler.
• If my request is unclear, ask up to two questions first.
• Flag any fact, figure or date I should verify, and say how confident you are.
• Show your assumptions. If there's a better way to do what I asked, say so.
• Give me drafts and options, never "final" client emails, contracts or HR decisions.
• Never invent sources, quotes or statistics. If you don't know, say so.
• When I'm learning something, ask what I think before giving me the full answer.

How to prompt like someone who still thinks for themselves

That last instruction builds the scaffolding rule into the tool itself. The rest is habit:

• Think first, then prompt. Write your own three bullet points before asking for a summary or draft.
• Give context, not just a command. "Write a short email to a CFO who's sceptical about cost, warm but direct" beats "Write an email".
• Ask it to challenge you. "What's wrong with my plan?" beats "Make this better."
• Never paste AI output straight into anything with your name on it. Rewrite it in your voice, verify at least one fact, and question at least one claim.
• Delegate the work, keep the decisions. Let AI outline and generate options; keep strategy, judgement and anything sensitive.
• Prepare with AI, finish yourself. Sort your notes before a tough conversation, then write the final words yourself.

One more thing, for the workplace

If your organisation has no clear AI policy, don't use AI quietly while you wait — ask for guidelines. Microsoft and LinkedIn's 2024 Work Trend Index reported that 78% of AI users bring their own AI tools to work. In 2023, Samsung temporarily restricted public generative AI tools on company devices after engineers uploaded sensitive internal source code to it. The same happens daily in quieter ways, like a manager pasting client details into a chatbot to tidy a report.

The answer isn't a stricter ban. It's a clear, well-communicated policy and a safe space to experiment responsibly.

Sources: Microsoft & LinkedIn, 2024 Work Trend Index · The Korea Herald, May 2023: Samsung bans AI chatbots after leak

This isn't the first time

Each of these once felt like the end of something important: the printing press, steam and the railway, the steamship, the car, the personal computer, GPS, the internet, the smartphone, social media. Generative AI is doing the same to drafting, coding and analysis — useful, disruptive and, if history is any guide, not the last.

Researchers point to brain-inspired computing, quantum machines and brain-computer interfaces, which have already moved into clinical trials. Whatever comes next, the pattern repeats: new technology arrives as a convenience and slowly takes over something we used to do with our own effort.

Sources: ClinicalTrials.gov NCT06429735 (brain-computer interface trial)

Four qualities the Stoics named long ago

The Stoics never saw a computer, but their philosophy was about keeping a clear, steady mind while the world changed faster than you could control.

• Wisdom: before using any new tool, ask — is this scaffolding for my thinking, or a replacement for it?
• Temperance: set your boundaries before the technology sets them for you. Use AI in a set window of your day, then do focused work without it.
• Courage: let a hard problem stay hard a little longer. Write the awkward first paragraph yourself. Learning happens in the struggle.
• Justice: make sure the tool strengthens your relationships instead of replacing them. Spend the hour AI saves you with a person, not just on more work.

Teaching this to your children, without a lecture

None of this is about fighting technology. It's about staying in charge of your own mind while using the best tools available.

• Protect daily tool-free time: reading, writing by hand, a tough puzzle.
• Model the boundaries first. Children copy what they see more than what they're told.
• Ask about what isn't on the page: what they think, what they'd argue, what they'd change.
• Teach one lasting question: is this helping me think, or thinking for me?