Every few months, a new AI tool launches with claims that it will revolutionize how we work. Most don't. Some do.

After testing dozens of tools and integrating AI deeply into my daily workflow, here's what actually stuck—and why.

The Core Stack

These are the tools I use almost every day:

1. Claude (via Claude Code CLI)

What it is: Anthropic's Claude model, accessed through the Claude Code command-line interface.

Why it's my primary tool:

  • Exceptional at complex reasoning and multi-step tasks
  • Understands large codebases when given proper context
  • Excellent at explaining why, not just what
  • The CLI integration means I never leave my terminal

What I use it for:

  • Writing and refactoring code
  • Debugging complex issues
  • Architectural discussions
  • Documentation generation
  • This entire website was built with Claude Code assistance

Limitations:

  • Can be overly cautious (sometimes refuses reasonable requests)
  • Context window, while large, still has limits
  • No real-time information

Cost: Claude Pro subscription + Claude Code access

2. ChatGPT (GPT-4)

What it is: OpenAI's flagship model, accessed via web interface and API.

Why I keep it around:

  • Better for creative and open-ended tasks
  • Broader general knowledge
  • Plugins and GPTs for specific use cases
  • Sometimes gives different perspectives than Claude

What I use it for:

  • Marketing copy and content ideas
  • Research on unfamiliar topics
  • Quick questions that don't need deep reasoning
  • Second opinions when Claude's answer seems off

When I choose ChatGPT over Claude:

  • Creative writing tasks
  • Brainstorming sessions
  • When I want a more conversational tone
  • Quick lookups and general questions

Cost: ChatGPT Plus subscription

3. GitHub Copilot

What it is: AI pair programmer integrated directly into VS Code.

Why it works:

  • Zero friction—suggestions appear as you type
  • Trained specifically on code
  • Understands file context automatically
  • Great for boilerplate and repetitive patterns

What I use it for:

  • Autocompleting obvious code
  • Writing similar functions based on existing patterns
  • Generating test cases
  • Filling in repetitive structures (imports, types, etc.)

Limitations:

  • Suggestions can be wrong in subtle ways
  • Sometimes completes in directions you didn't want
  • Less useful for novel or complex logic

Pro tip: Learn the keyboard shortcuts. Tab to accept, Esc to dismiss, Ctrl+Enter to see alternatives. Speed matters for flow.

Cost: GitHub Copilot subscription

The Supporting Cast

Tools I use regularly but not daily:

Cursor

What it is: VS Code fork with deep AI integration.

My take: Impressive capabilities, but I found myself preferring the Claude Code CLI workflow. Cursor shines for developers who want AI embedded in their editor. I prefer keeping AI in the terminal.

Best for: Developers who want inline AI assistance without leaving their editor.

Perplexity

What it is: AI-powered search with citations.

Why it's useful: When I need factual information with sources. Better than asking ChatGPT when accuracy matters because it shows where information came from.

What I use it for:

  • Technical research with citations
  • Checking current information
  • Comparison shopping for tools/services

v0 by Vercel

What it is: AI that generates React/Next.js UI components from descriptions.

My take: Genuinely useful for prototyping UI. I describe what I want, it generates a starting point. Not production-ready code, but excellent for getting 70% of the way there quickly.

What I Stopped Using

Just as important as what works is what doesn't:

Generic "AI Coding Assistants"

Many tools promise to "write code for you" but lack the context awareness to be useful. If it can't understand my codebase, it can't help with my codebase.

AI-Powered IDEs I Don't Already Use

Switching editors has a high cost. Unless the AI integration is dramatically better (it usually isn't), I'd rather use tools that integrate with my existing setup.

"One-Click" App Builders

Tools that promise to build entire apps from a description. In my experience, they create demos, not products. The gap between "looks like it works" and "actually works" is where all the real work lives.

How I Choose Tools

My criteria for adopting a new AI tool:

1. Does It Fit My Workflow?

The best tool that breaks my flow is worse than a decent tool that integrates smoothly. Claude Code works because it's in my terminal where I already live.

2. What's the Learning Curve vs. Payoff?

Some tools require significant investment to use well. That's fine if the payoff justifies it. Many don't.

3. Can I Trust the Output?

AI that's right 95% of the time sounds good until you realize you have to check everything anyway. I prefer tools where I can quickly verify correctness.

4. What's the Failure Mode?

When the tool fails (and it will), how bad is it? Copilot suggesting wrong code is easy to spot. An AI architect making bad design decisions might not be obvious until later.

Cost Analysis

Here's what I actually pay monthly:

| Tool | Cost | Value Assessment | |------|------|------------------| | Claude Pro | $20/mo | Essential—primary development tool | | ChatGPT Plus | $20/mo | Useful—different perspective, creative tasks | | GitHub Copilot | $10/mo | High value—constant small time savings | | Perplexity Pro | $20/mo | Nice to have—could use free tier | | Total | ~$70/mo | |

Is $70/month worth it? For me, absolutely. If AI saves me 5 hours per month (conservative estimate), that's $14/hour for tools that actually save me 20+ hours.

The key is being honest about which tools you actually use. Don't pay for tools that sit idle.

Recommendations by Use Case

If you're just starting with AI development:

  • Start with ChatGPT Plus (lowest friction, broadest capability)
  • Add GitHub Copilot once you're comfortable

If you're a serious developer:

  • Claude (via Claude Code) for complex tasks
  • GitHub Copilot for in-editor assistance
  • ChatGPT for variety and creative tasks

If you're budget-conscious:

  • Free tiers of Claude and ChatGPT go surprisingly far
  • GitHub Copilot has the best bang-for-buck at $10/month

If you're building a product:

  • API access to Claude or GPT-4 for integration
  • Consider costs at scale—tokens add up

The Meta-Lesson

The most important thing isn't which specific tools you use—it's developing the skill to use AI tools effectively.

Prompt engineering, understanding model capabilities, knowing when to trust and when to verify—these skills transfer across tools. Today's best model might be tomorrow's second choice, but your ability to work with AI compounds over time.

Next step: Learn how to get better results from whatever tools you choose: Prompt Engineering for Code Generation

This post is part of the AI & Automation series. For the big picture, see AI-Assisted Development: A Solo Founder's Toolkit.