My favourite AI trick isn't a fancy tool (although there are many cool ones) or a long and convoluted prompt that you need to comment “Prompt” into an Instagram post to access. It's using AI to fix the way I'm using AI.

(And yes, I appreciate how insufferably meta that sounds.)

Here’s what it looks like in practice.

When I’m not sure how to structure something

Last week, I wanted to work with AI to help me rebuild Inventium’s self-reflection process (our version of performance management / development). I had a vision of what I wanted to achieve, but I wasn’t sure where exactly to start. So I asked Claude (my AI tool of choice) to map out a structure with me first. Then I used that map as the foundation for building the actual thing.

The prompt I use:

“I want to build an [X] that does [Y]. I’m not sure how to structure this for an AI to execute. Can you interview me about what I need, then map out the steps in order?”

When I need prompts for other AI tools

Image generation tools like Nano Banana work much better with detailed, well-structured prompts. Rather than writing those myself, I have Claude create the first draft based on guidelines I provide (e.g. must be photorealistic; must use Inventium brand colours), and I refine the guidelines as the outputs improve.

The prompt:

“Here are my guidelines for what I want: [paste guidelines]. Write 3 detailed image prompts based on these. After I give you feedback, update the guidelines so future prompts get better.”

When I don’t fully know what I need yet

Lately, I’ve been building Skills in Claude to help me with various (and very specific) tasks. Occasionally, I have a sense of what I need but I’m not sure on the details. So I ask Claude to interview me before we start. A round of good questions surfaces details I would have missed, and the answers become the brief.

The prompt:

“Before we start, I want you to interview me about [X] to create [Y] so we don’t miss anything important. Ask me the questions you’d need answered to do this well.”

When AI output keeps missing the mark

Instead of rewriting the output myself and moving on, I’ll show Claude the draft and my edited version and ask it to figure out what changed and why. That diagnosis becomes the updated prompt for next time.

Prompt:

“Here’s the prompt I used: [paste]. Here’s what it produced: [paste]. Here’s my edited version: [paste]. What did I change and why? Use that to rewrite the original prompt so future outputs need less fixing.”

When I’m building a prompt and need some help

If I need a complex prompt but I’m not sure what it should contain, I describe the job to Claude and ask it to write the prompt for me. Then I critique the prompt rather than starting from scratch.

Prompt:

“I need a prompt that will reliably get Claude to do [X]. The output should [describe]. The likely inputs will be [describe]. Write me a draft prompt, then explain the choices you made so I can refine it.”

When I want consistent outputs across a team

If multiple people are using AI for the same task and getting vastly different results, I’ll ask Claude to help me build a repeatable template (inputs, constraints, output format) that anyone can use without needing to think too hard about prompting.

Prompt:

“I need a reusable AI template for [task] that anyone on my team can use. The typical inputs will be [describe]. The output should always look like [describe]. Build me a fill-in-the-blanks prompt template with instructions for how to use it.”

When I’m not sure if my prompt is the problem

Before scrapping an approach entirely, I’ll ask Claude to audit the prompt I’ve been using and tell me exactly what’s ambiguous, missing, or likely to produce inconsistent results.

Prompt:

“Here’s a prompt I’ve been using: [paste]. I’m not getting consistent results. Audit this prompt. Tell me what’s ambiguous, what’s missing, and what would you change to make outputs more reliable?”

Mediocre outputs are almost never an AI problem. They're a "I didn't think this through properly before I started" problem. Which is, as you can see above, very fixable.

Let me know in the comments: which prompt will you try today?