How to Use AI Effectively: A Builder's Framework

Most people use AI like a search engine with better grammar. That's about 10% of what it can do. Here's the framework I use to get real work done with AI every day.

The Three Modes

Mode 1: AI as Assistant

You drive, AI helps. Good for:

• Drafting emails and documents
• Answering specific questions
• Brainstorming ideas
• Quick code snippets

How to use it well: Be specific. "Write a cold email to SaaS founders about our analytics tool, emphasizing speed of setup" beats "write a marketing email."

Mode 2: AI as Agent

AI drives, you review. Good for:

• Building full features (Claude Code)
• Writing complete articles
• Analyzing data sets
• Code review and testing

How to use it well: Define the goal and constraints clearly. Let the agent iterate. Review the outcome, not every step. This is vibe coding applied to all work.

Mode 3: AI as System

AI runs continuously without you. Good for:

• Automated code review (GitHub Actions)
• Customer support triage
• Content scheduling
• Monitoring and alerting

How to use it well: Build once, monitor weekly. Set clear boundaries for when to escalate to human attention.

The Prompting Framework

Good AI output starts with good input. My formula:

Context + Task + Format + Constraints

• Context: What's the situation? What do you know?
• Task: What exactly do you want?
• Format: How should the output look?
• Constraints: What should it avoid or include?

Example: "I'm building a Next.js SaaS app (context). Add a Stripe checkout flow to the pricing page (task). Use server actions, not API routes (format). Don't modify the existing layout or navigation (constraints)."

Tool Selection

Different tools for different work:

| Work Type | Best Tool | |-----------|-----------| | Software development | Claude Code | | Quick code edits | Cursor | | Writing & research | Claude.ai or ChatGPT | | Data analysis | Claude with file upload | | Design | v0 + AI image tools | | Automation | AI workflow tools |

Don't use one tool for everything. Match the tool to the task.

Common Mistakes

Being too vague. "Make it better" gives you generic output. "Reduce the function from 40 lines to 15 by extracting the validation logic" gives you exactly what you need.

Not reviewing output. AI is confident, not always correct. Always verify facts, logic, and code before using.

Fighting the tool. If AI can't do something after 3 attempts, try a different approach or a different tool. Don't keep rephrasing the same broken prompt.

Ignoring context. AI works better with more context. Share your codebase, your goals, your constraints. The more it knows, the better it performs.

Using AI for the wrong tasks. AI excels at well-defined, pattern-based work. It struggles with novel creative decisions, emotional intelligence, and ambiguous goals.

Building AI Into Your Workflow

1. Start with one task. Pick the thing you do most that AI could help with.
2. Master the tool. Spend a week getting good at one AI tool.
3. Expand systematically. Add one new AI-assisted workflow per week.
4. Automate what works. Turn repeated AI interactions into automated systems.

For the complete business application, see One-Person AI Business.

Frequently Asked Questions

What's the single most important tip for using AI effectively?

How much time should I spend learning AI tools?

Is AI going to replace my job?