Hints and Tips

Little habits that make working with AI smoother.

Let the AI ask you the questions

When you've got something fuzzy to work out — a decision, a plan, the shape of a new project — you don't have to write it all up first. Flip it around: ask the AI to interview you.

Instead of staring at a blank page trying to write an essay, you just answer simple questions as they come, and the thinking gets done for you. The "one at a time" part matters — without it you'll get ten questions at once and be right back to writing an essay.

How the AI's memory works (context windows)

Every AI conversation has a memory limit. The proper name for it is the context window, and the things that fill it up are called tokens (roughly: words and bits of words — everything you type and everything the AI says back).

Three things are worth knowing:

• Long conversations get slower and cost more. The AI re-reads the whole conversation every time it replies, so the longer it gets, the more work each reply is.
• Eventually it starts to forget the beginning. When a chat gets really long, the earliest bits fall out of memory. It won't announce this; things just quietly get a bit off — it forgets a detail you set up an hour ago, or repeats itself.
• You can check how full it is. Type /context and Claude shows you what's taking up the space and how much room is left.

Want it on screen the whole time? You can add a little status line along the bottom of the window that always shows how full the context is, as a percentage — so you can just glance at it instead of asking. (This is the bar some people have running at the bottom of their Claude window.) It's a one-time setup that Claude can do for you:

How full is too full? There's no exact line, but once it climbs past half, the AI's sharpness gradually starts to slip — so aim to finish your current chunk and start fresh rather than pushing on if you're anywhere near there.

So when a session feels slow or the AI seems to have lost the plot, that's the context window filling up — and the fix is the next section.

Do a chunk of work, then start fresh

The simplest habit for keeping the AI sharp: do a chunk of work, then clear and start a fresh conversation. A "chunk" is whatever naturally hangs together — that might be one job, or two or three small ones in a row. You don't have to start over after every tiny task.

To wipe the slate and begin clean, type:

This empties the conversation's memory and gives you a fresh, fast AI. Your files are untouched — /clear only clears the chat, not your work.

Picking up where you left off

When a conversation fills up, you've really got two choices, and it's worth knowing the difference:

• Compact it. You can ask Claude to compact the conversation — it squashes everything so far into a short summary and keeps going from there. Type /compact. You keep the thread of what you were doing, but you lose the fine detail (it's working from the summary now, not the full history).
• Start fresh. Clear it (/clear) or open a brand-new conversation. Sharpest and fastest, but the AI remembers nothing — so you re-brief it, like talking to a stranger: "I'm building X. Yesterday we did Y. Today I want to do Z."

Rule of thumb: compact when you're mid-thought and want to keep momentum; start fresh when you're beginning a new chunk of work and want the AI at its sharpest. Either way, a plan or notes file it can re-read is the safety net that means nothing important gets lost.

Leave yourself notes

Because the AI forgets when you clear or start fresh, it's worth parking things outside the conversation, where they survive.

Mid-task thought you don't want to chase right now? Just ask the AI to note it down:

And you can leave notes for future you the same way: a line or two about where you got to and what's next, so tomorrow's fresh conversation can read it and catch up instantly.

Ask another AI to help you write prompts

If you're not sure how to ask Claude for something, ask a different AI to write the prompt for you. Open claude.ai (or Gemini, or ChatGPT) in your browser and say what you're trying to do — it'll hand you a well-worded prompt you can paste straight into your tutor or your project.

It feels a bit like cheating. It isn't — getting AIs to help you talk to other AIs is a perfectly good move.

It's okay to say "please do it for me"

Sometimes Claude will ask you to do a step — click something, paste something, go and check a setting. You don't always have to. Often you can just hand it back:

If it genuinely can't (some things really do need a human — signing into a website, clicking a button in your browser), it'll tell you so and explain why. But a lot of the time it asked out of politeness, not necessity — and it's happy to take the wheel.

Getting on with other things while it works

Some jobs take Claude a minute or two. Tempting to open a second Claude and do more at once — but be careful: two Claudes working on the same project at the same time will get in each other's way and tangle things up. One project, one Claude at a time.

A nicer trick: get Claude to make a sound when it's finished, so you can go and do something else on your machine and come back when you hear it. This kind of set-it-once automation is called a hook — a small instruction that runs by itself whenever a particular thing happens (here, whenever Claude finishes its turn). You set it up once and it works every time after:

You can put the laptop to sleep

You don't have to finish everything in one sitting. You can shut the lid, put the laptop to sleep, and come back later — Claude will pick up right where you left off.

One small thing: when you wake the laptop back up, give it a few seconds before you start typing. It needs a moment to reconnect to the internet in the background. Type too soon and you might get a hiccup; wait a beat and it's fine.

Make your terminal dark — or light

Prefer a dark, night-mode look (or the other way round)? The background colour is set by your terminal program itself, not by Claude — which is why Claude's own theme picker doesn't change it. Here's where the real switch lives:

• On Windows: it depends on what your PowerShell window looks like (this varies by Windows version — no need to change anything). Tabs along the top? Open the ⌄ menu next to the tabs → Settings → Appearance → Color scheme. Plain window, no tabs? Right-click the title bar → Properties → Colors.
• On a Mac (Terminal): menu bar → Terminal → Settings → Profiles. Pick a profile from the list on the left — "Basic" is the light one you start with, "Pro" is a black one — then click Default at the bottom so it sticks next time.
• On Linux: terminal programs vary, but look in your terminal's menu (often a ☰ button, or right-click inside the window) for Preferences, then a Colours or Profiles section — the background and text colours live there.

It's purely cosmetic and completely reversible, so have a play until it's comfortable on your eyes.

Useful slash commands

"Slash commands" are special commands you type to the tutor that start with a slash. The tutor recognises these and acts on them instead of treating them as a normal message.

You hardly ever need these; the tutor handles most things in chat. But these few are worth knowing.

/help: show the full list of commands the tutor knows.

/clear: start a fresh conversation — the AI forgets the current chat. Your saved progress in progress.md is safe, and the old conversation is still there under /resume if you need it.

/exit: leave the tutor cleanly. You can also just close the terminal window; your progress is saved as you go, so nothing is lost.

For undoing AI changes: just ask the tutor in chat: "undo that" or "can you take us back to before the last change?" The tutor uses git to roll you back to the last save-point. There's no special command to remember; the conversation is the controls.