AI coding tools

Almost every AI tool fits into just five core categories that haven't changed in years.

Most people use AI thinking tools like Google, asking a question and getting an answer, and that's the entire relationship.

When AI is used only as a Q&A tool, it feels overhyped.

You should stop asking AI questions and instead make it build things for you.

The single best way to build with AI is to make the AI write its own prompts because it knows what it needs better than you do.

The software building category of AI tools would have required a developer and a few thousand dollars five years ago.

Now you just describe what you want, the AI writes all the code and does the hard work, and you only need to be creative.

Software building tools split into two types: 'describe it and it appears' tools (Lovable, Bolt, Base44) and agentic coding tools (Claude Code, Cursor, Codecs).

For more than one or two projects, the agentic route (Claude Code, Cursor) almost always comes out cheaper than credit-based 'describe it' tools.

Before building anything with AI, you must make a plan (e.g., using plan mode) to reduce mistakes.

Always keep your project's core context in a markdown file (e.g., claude.md, gemini.md, agents.md) so the AI always has the necessary knowledge.

With agentic coding tools, you must watch security: stay in the loop, review actions before approving, and never paste real passwords or API keys.

Research and knowledge tools fix the single biggest weakness in AI: generalist chatbots cannot beat specialist tools that read real, current sources with clear provenance.

The three research/knowledge tools actually worth using are Perplexity, Deep Research, and NotebookLM.

The rule for using research tools is: whenever an answer needs to be true, current, or specific, you should use one of them.

AI image creation tools are effectively a free design department that can produce logos, icons, product mockups, and thumbnails in seconds.

AI image tools won't fully replace designers, but for quick tasks like the example shown, they are 100% worth it.

Most people get poor AI image results because they don't know how to prompt; vague requests like 'a logo for my app' give the AI too little to work with.

Better image prompts are extremely descriptive yet focused, and specificity is the key; you can also have a chatbot sharpen your prompt beforehand.

Instead of generating from scratch, you can now provide a reference image and edit results with plain English instructions.

One of the easiest ways to get high-quality (though sometimes limited) free AI images is to use Gemini's built-in image generation.

AI Studio gives access to Nano Banana 2, one of the best image models, but it is paid via API key and pay-as-you-go.

For AI video, always start from an image rather than generating video from nothing; this gives more control, consistency, and reduces face-drifting.

High-quality AI video requires keeping motion small and short because large movements tend to look unrealistic.

Gemini's video generation requires a subscription, and most AI video tools are paid because AI video is expensive for companies to run.

Once you understand how to use the five AI categories, you can build practically anything.