AI coding tools

A year ago, getting real results from AI was hard, but that has completely changed recently.

The biggest mistake people make is constantly switching between AI tools before learning what any one of them can actually do, rather than choosing the 'wrong' tool.

A beginner who learns Claude deeply will be far ahead of someone bouncing between ChatGPT, Claude, Gemini, Perplexity, and five other tools at a shallow level.

The three ecosystems worth choosing from are ChatGPT, Claude, and Gemini; just pick one of those three.

The platforms have only become good enough and easy enough for anyone to learn in the past few months.

Complex prompt engineering is usually a waste of time; instead, use simple ICC (instructions, context, constraints).

Using a 'context interview' where you ask the AI to request more context before answering leads to dramatically better, more personalized responses.

Giving more context is usually better; these models can handle a full book of context without being overwhelmed, as long as it is relevant.

One should treat the first AI output as a draft, iterate until perfect, and treat the tool as a collaborator.

AI models hallucinate, but you can reduce or catch hallucinations by asking for confidence levels, citing sources, finding opposing experts, or getting a critique from a different model.

Setting up a project within your chosen ecosystem, with custom instructions, memory, and a knowledge base, creates a persistent AI expert for each topic, dramatically improving results.

After setting up a project, you can do a context interview and save the output as a document in the knowledge base, so the AI retains that context permanently.

Packaging a process as a 'skill' (in Claude) or equivalent (custom GPTs in ChatGPT, Gems in Gemini) allows the AI to repeat that process automatically in the future, saving significant time.

Using the same ecosystem regularly gets better over time because the system retains and adapts to your instructions and feedback.

Artifacts (Claude) or Canvas (ChatGPT/Gemini) allow you to build standalone interactive content like dashboards, web pages, or games within the chat interface using natural language.

Vibe coding—describing a tool in plain language and letting AI write all the code—allows non-technical users to create useful personal tools, connecting to existing apps and services.

Claude Code and OpenAI Codex are powerful for building tools, but Codex offers higher usage limits and is available on ChatGPT's free plan, making it more accessible for frequent building.

Google's updated Anti-Gravity is now competitive with Claude Code and Codex, though the author has limited experience with the latest updates.

To get maximum value from any ecosystem, one should upgrade to the paid plan; the free plans are insufficient for full capability.

ChatGPT is the broadest all-in-one AI app with great image generation and voice mode; Claude excels in writing; Gemini excels in Google ecosystem integration and YouTube analysis.

For image and video generation, Higsfield consolidates leading models in one place, making it more efficient than using multiple separate platforms.

Granola is a superior meeting notes tool because it does not require a bot to join the meeting and can connect to Claude for further processing.

Whisper Flow is a speech-to-text dictation tool that automatically corrects misspeaks, making it easier than traditional dictation.

N8 is powerful for deep customized agentic workflows but is significantly more technical to learn than anything else, including vibe coding tools.

Notebook LM is the best way to learn and organize information, grounding answers in user-provided sources and citing them, which reduces hallucinations.

Most other AI tools can be replaced by built-in ecosystem features or custom vibe-coded solutions, reducing the need for separate subscriptions.

Implementation, not just learning, is the key to getting results from AI; one should immediately apply a skill after watching a tutorial.