How to actually use AI tools well (without being intimidated)

The mental model that changes everything

Before any specific technique, change how you think about these tools.

Most people use AI like Google. They type a question, get a response, accept or reject it, move on. This treats the AI as an answer machine.

The shift that makes AI tools genuinely useful: treat them as a thinking partner instead. The AI isn't a vending machine that dispenses answers. It's a collaborator that can help you think through problems, generate options you wouldn't have considered, identify weaknesses in your reasoning, and accelerate the parts of your work that don't require your specific expertise.

The difference is everything. A vending machine relationship produces generic output. A thinking partnership produces work that's better than what either you or the AI could produce alone.

The four-part formula for good prompts

When you're asking an AI for help, the prompts that work best usually include four things:

Context. What's the situation? Who am I? What am I working on? What's the background?

Task. What specifically do I want you to do?

Constraints. What are the limits, requirements, preferences? Length, tone, format, what to avoid.

Examples. When possible, what does good output look like?

Compare these two prompts for the same task:

Bad prompt: "Write a cold email."

Good prompt: "I'm a fractional CFO reaching out to mid-size SaaS companies (50-200 employees) to offer my services. Write a cold email that's 80-120 words, professional but not stiff, ending with a specific call to action (booking a 15-minute discovery call). Avoid generic 'I hope this email finds you well' openings. Lead with a specific business observation that demonstrates expertise."

The second prompt produces dramatically better output because it gives the AI everything it needs to make good decisions. The first prompt forces the AI to guess at all the details.

Iterate, don't accept

The single biggest mistake people make: accepting the first response.

LLMs are good at producing useful output, but the first response is usually not the best. It's a starting point. The next steps are where the real value comes from.

After the first response, ask:

• "Can you make this 30% shorter?"
• "Rewrite this in a more conversational tone."
• "Give me three different versions with different angles."
• "What are the weaknesses in this approach?"
• "What am I missing?"
• "What would a critic say about this?"

The conversation is where the work happens. People who use AI well treat it as a back-and-forth, not a one-shot oracle. The third or fourth iteration is almost always better than the first.

Use AI to think, not just to produce

The deepest value of AI tools isn't producing finished work. It's helping you think.

Some prompts that change how I work:

"I'm thinking about doing X. What are the considerations I should be thinking about that I haven't mentioned?" This surfaces blind spots. The AI has been trained on huge amounts of human writing about most topics, which means it can often name considerations you haven't thought of.

"Here's my plan. What's the strongest argument against it?" Force the AI to steelman the opposing view. This is more useful than asking for general criticism, which often produces mild objections.

"Explain this to me like I know nothing about [topic]." Useful for genuinely understanding things outside your expertise. Then ask follow-up questions until the explanation actually clicks.

"What do experts in [field] disagree about regarding this question?" Helps you understand the actual state of knowledge rather than the simplified version most introductions present.

"What questions should I be asking that I haven't thought to ask?" Particularly useful at the start of a new project. Surfaces important questions you'd otherwise miss.

These prompts use the AI to extend your own thinking rather than to replace it. The output is intellectual support, not finished product.

Pick the right tool for the task

The major AI tools have different strengths. Knowing which to use for what saves time and produces better results.

Claude (Anthropic): My primary tool for editorial writing, analysis, code, and tasks requiring careful reasoning. The model is willing to admit uncertainty, push back on flawed premises, and engage substantively with complex topics. Particularly strong for longform writing and code that needs to be genuinely good rather than just plausible.

ChatGPT (OpenAI): The most widely-used option, with the most third-party integrations and plugins. Good for general use, strong for creative tasks. The DALL-E integration is convenient if you need images. The custom GPTs ecosystem can be useful for specialized tasks.

Gemini (Google): Strongest when you're already deep in Google's ecosystem (Docs, Sheets, Workspace). Useful for tasks involving Google services directly. Less differentiated for general-purpose work.

Grok (xAI): Integrated with X, useful for tasks involving real-time social media context. Less restrictive content moderation than other major models, which some users value and some find off-putting.

Perplexity: Different category. Optimized for research with real-time web search and citations. Useful when you need current information or want sources you can verify.

For most people, picking one primary tool and learning it well beats hopping between tools. Add others only when you have a specific need they handle better.

What to delegate vs what to keep yourself

Not every task should go to an AI. The pattern that works:

Delegate the formulaic parts. Drafts of standard emails, summaries of documents you've already read, first-pass code for well-defined functions, formatting work, repetitive tasks. The AI is faster than you at these.

Keep the strategic parts. What you're actually trying to achieve, what your perspective is on the situation, what feels right based on your specific knowledge. The AI doesn't have your context, relationships, or judgment.

Use AI as a force multiplier on your thinking. Once you've decided what to write, the AI can help you draft. Once you've decided what argument to make, the AI can help you structure it. Once you've decided what code to build, the AI can accelerate the implementation. The decisions stay yours. The execution accelerates.

The failure mode: trying to delegate decisions you should be making yourself. "Should I take this job offer?" is not an AI question. "What questions should I be asking myself before deciding whether to take this job offer?" is.

Verify, verify, verify

LLMs are confidently wrong sometimes. The verification habits matter.

For facts: Don't trust specific dates, numbers, statistics, quotes, citations, or names without verification. The AI can produce plausible-sounding fabrications. Use the AI to draft, then verify the specifics with primary sources.

For code: Run it. Test it. Read it. Don't assume working code is correct code. AI-generated code can have subtle bugs that look fine on inspection.

For analysis: Check the reasoning. The AI can produce arguments that sound right but rest on hidden assumptions. Read your AI's output critically the same way you'd read a colleague's draft.

For anything that matters: If the cost of being wrong is high, verify everything. If the cost is low, the AI's first draft is usually fine.

The general rule: AI is great for accelerating your work, dangerous for replacing your judgment.

Specific workflows that work

A few patterns I use regularly:

The brainstorm-and-prune pattern. Ask the AI to generate 10-20 options for something. Then pick the 2-3 that resonate and develop those further. The AI is better at generating quantity. You're better at picking what's actually good.

The draft-and-rewrite pattern. Have the AI produce a first draft. Then rewrite it in your own voice, keeping the structure but making the language yours. This is faster than writing from scratch and produces output that sounds like you, not like an AI.

The explain-and-question pattern. Have the AI explain something to you. Then ask follow-up questions about the parts that didn't make sense. Keep asking until you understand. Better than reading a textbook for getting an intuition.

The critique-my-thinking pattern. Write out your reasoning on a decision. Paste it to the AI. Ask "what's wrong with this thinking?" This is how I catch my own blind spots before they cost me.

The summarize-this-document pattern. Paste a long document. Ask for a summary with specific structure (key points, action items, open questions). Useful for quickly extracting value from documents you don't have time to read fully.

What not to do

A few patterns that don't work:

Don't paste in confidential information. Most AI tools retain some version of your inputs for training or quality improvement. Don't paste in client confidential data, proprietary code, or anything you wouldn't want to potentially leak. The major models have enterprise versions with stricter privacy guarantees if you need to work with sensitive information.

Don't ask AI to "be" you on important communication. AI-drafted personal messages, breakup texts, condolence notes, important professional communications. People can tell. The cost when they realize is high. Use AI to brainstorm or polish, but write the actual message yourself when relationships are involved.

Don't use AI for decisions you should be making yourself. Career choices, relationship decisions, big life questions. The AI doesn't know you well enough, doesn't have skin in the game, and won't be there when the decision plays out.

Don't blindly trust AI for things you'd verify with a human. If you'd get a second opinion from a doctor, lawyer, accountant, or financial advisor, get a human opinion. The AI is a starting point, not a substitute for professional judgment on high-stakes questions.

The compounding skill

Using AI well is itself a skill, and the skill compounds over time.

The first month, you'll get decent output. The third month, you'll get noticeably better output because you're learning what prompts work. The first year, you'll be producing work that you genuinely couldn't have produced without these tools. The compounding is real.

The people who will have the biggest advantages over the next decade aren't the people who are smartest. They're the people who learn to pair their human judgment with AI capability the most effectively. That skill is available to anyone who's willing to actually practice it instead of just reading tutorials about it.