AI for beginners

I found myself repeating these five tidbits whenever friends getting started with AI asked me for advice. Hope they save you some time, and feel free to pass them on to pals just getting started.

1. Give it loads of context (using a voice prompt)

Most people ask AI a short question and get a generic answer.

Instead, give it far more context than feels necessary.

If you are planning a trip, do not just say:

“Plan me a week in Japan.”

Tell it what you liked last time, what you hated, your budget, who you are travelling with, how fast you like to move, and anything else that might matter.

You do not need to write a perfect prompt. Ramble.

AI is good at handling spelling mistakes, tangents, unfinished thoughts, and messy context.

I use Handy on macOS for this. I tap a shortcut to start recording, talk for a minute or two, and drop the transcription straight into ChatGPT, Claude, or Perplexity.

The more real context you give it, the less it has to guess.

2. Understand sycophancy

AI has a tendency to agree with you.

You might suggest an idea and immediately hear:

“That’s a great idea.”

Maybe it is. But AI can inherit your framing, reward your confidence, and make weak ideas sound stronger than they are.

Sometimes you need encouragement. Other times you need an objective answer.

Do not mistake enthusiasm for objectivity.

3. Red-team the plan

Once you have a plan you like, ask AI to attack it.

The term “red team” comes from the military, where one side plays the opposition and tries to find weaknesses in the other side’s plan.

You can simply say:

“Red-team this.”

Or, more directly:

“Stop helping me prove this is a good idea. Try to prove it is a bad one.”

Ask it what is weak, missing, overly optimistic, or likely to go wrong.

This is one of the best filters for AI’s natural tendency to agree with you.

4. Ask another AI

For bigger decisions, I often copy the finished plan into another AI and say:

“A friend gave me this plan. What do you think?”

That gives you a fresh perspective without all the momentum and context of the original conversation.

Ask what is missing, what could go wrong, and which assumptions look weak.

Then paste that critique back into the original chat and let it respond.

One or two loops like this can noticeably improve a plan.

5. Reverse-engineer what worked

Sometimes it takes ten messages to finally get the answer you wanted.

When you get there, do not just leave.

Ask:

“What should I have asked you at the beginning to get much closer to this result?”

Then save that prompt.

This is one of the easiest ways to get better at prompting because you are learning from your own real examples.

The common thread

Do not treat the first AI response as the answer.

Give it context. Know that it may be trying to please you. Challenge its recommendations. Get a second perspective. Then learn from the conversations that worked.

Treat the first answer as a draft, the second as a critique, and the real world as the final check.