This week, we’re taking three small steps beyond basic AI: giving it the context to understand you, learning why direct instructions work better than politeness, and discovering that you can build genuinely useful tools without being a programmer. I’m hoping these “next steps” are helpful in your AI journey; I had fun with them.
What this is
TechCurious This Week is a short bi-weekly briefing on the news and latest tips and advances in AI, how it can help you, what it means for your life, and what, if anything, to do about it. Three stories or ideas, decoded, with my honest read. The standing deal: if something matters, I’ll tell you why. If it doesn’t, I’ll tell you that too — some weeks the most useful thing I can say is ignore this one.
1. The three-minute setup that makes every answer better
If you regularly have to remind an AI that you’re in Canada, that you want plain language, that you dislike long introductions or that you don’t always want to be praised, you’re doing work the tool can remember for you.
Most major AI platforms have a place for standing instructions: a short description of who you are, what you use the tool for and how you want it to respond. This matters because AI fills in blanks. If you don’t give it useful context, it guesses — and you get answers that are generic, too wordy, too enthusiastic or aimed at somebody else.
A useful standing brief can be very simple:
I run a small business and I’m short on time. Use plain language, be direct, and explain unfamiliar terms. Give me the answer first and the detail second. Don’t use jargon, excessive praise or exclamation marks. If my idea has a weakness or risks, tell me.
Here’s where to put it:
ChatGPT
- On a computer (web or desktop app): Click your profile icon, then Settings → Personalization → Custom Instructions. Turn on Enable customization and add your instructions.
- On iPhone or Android: Tap your profile icon to open Settings, at the top is Customize ChatGPT, click Personalization, then add your instructions in the Custom Instructions field.
- As of July 15, Plus, Pro, Enterprise, Business and Education users have up to 5,000 characters instead of 1,500, so you can now give it a more complete brief. ChatGPT’s Memory is separate: Custom Instructions are what you deliberately tell it; Memory is what it learns from your conversations. On a computer, review Memory under Settings → Personalization → Memory. OpenAI explains Custom Instructions and the distinction from Memory.
Claude
- On a computer (web or desktop app): Click your initials in the lower-left corner, choose Settings, go to General, then enter your account-wide preferences under Instructions for Claude.
- On iPhone or Android: Open the left sidebar, tap your initials or name to open Settings, click Profile, then fill in your information under Instructions.
- If you use a Claude Project, add separate project instructions for that specific work. Claude’s personalization guide.
You don’t need to write your autobiography. Start with four things: who you are, what you commonly do, how you want answers structured, and what annoys you. Then revise it when the results aren’t quite right. Three minutes now can save you from reintroducing yourself in every new conversation.
2. Please stop saying please. (Sort of.)
Have you noticed that if you say “please” and “thank you” to ChatGPT, it often becomes even more polite in return? It praises the question, reassures you that you’re on the right track and wraps the actual answer in several layers of conversational bubble wrap.
The internet’s current solution is: be rude to it. That advice came partly from a small Penn State study in which blunt or rude versions of 50 questions performed a few points better than very polite ones. It made a terrific headline. It did not settle the question. An earlier study found that impolite prompts could actually make results worse, and performance varied by model and language.
The useful lesson isn’t “be mean.” It’s stop making the AI hunt for the request inside all your softening language.
Instead of:
If it isn’t too much trouble, could you possibly help me think through whether this might be a good idea?
Try:
Assess this idea. Tell me what is strong, what is weak and what I may be overlooking. Be direct.
“Please” isn’t breaking ChatGPT, and good manners are a perfectly reasonable habit. But AI doesn’t need the social cushioning we use with people. It works better when the task, audience, format and constraints are obvious.
For many women, that may be the more interesting experiment: not learning to be rude, but practising saying exactly what we want without apologizing for the ask. ChatGPT will cope.
3. Ready to go a little deeper? Here’s where to start.
If you’ve been talking to AI for months, the next step may be to stop asking it only for answers and ask it to make something. I’ve started to write about this in my Beyond ChatGPT Series and really liked the specific advice I read last week. Jenny Ouyang’s Make the Most of Claude Code is a useful on-ramp because it doesn’t begin with a grand app idea. It starts with four small projects based on files you already have:
- A one-page personal website built from your résumé, biography or business notes.
- A visual expense tracker that turns sample transaction data into categories and charts.
- A daily inspiration tool that draws from your own saved quotes, notes or journal entries.
- A photo-organizing helper that proposes clearer filenames and sorts copies into folders by date.
Are you ready to try “vibe coding?” – you describe what you want in ordinary language, the AI writes the code, and you test, correct and improve what it produces. You are still making the decisions; the machine is handling much of the technical translation.
Each idea above introduces the same basic building loop:
- Choose one small outcome you can describe clearly.
- Put the relevant material in a new folder — use copies or sample data for your first attempt, not irreplaceable photos or sensitive financial files.
- Open Claude Code in that folder and paste the suggested starting prompt.
- Ask it to explain its plan and show you a preview before it changes anything. Test the result, say what isn’t right, and let it revise.
The real goal isn’t the website or the photo folders. It’s the moment you realize that you can describe a useful tool, watch it take shape and improve it through conversation. That’s a very different relationship with AI than asking it to rewrite an email — and it’s where things start to get interesting.
P.S. This does require Claude Code and some setup, but give yourself some time and follow the instructions in the link above and the beginner guide to Claude Code. If you get stuck, remember Claude is your thinking partner - just ask it: “what does that mean”, “what do I do next” or even “can you do that for me?”
That’s this week’s edition: give it context, say what you mean and, when you’re ready, ask it to build.
Consider it decoded.
— Amanda