AI Coding

I Thought Codex Could Build a Mahjong Game If I Just Asked

I wanted a free browser mahjong game for friends, so I asked Codex to help build one. The project worked enough to publish, but mahjong exposed the real limit of AI coding: human understanding still matters.

How to Use Codex Practically Before Mobile Support Reaches Everyone

OpenAI has brought Codex into the ChatGPT mobile app in preview. Here is how beginners can get a first real Codex win with HTML, JavaScript, Python, and a realistic PC plus phone workflow.

Did Claude 4.7 Make Japanese More Expensive? Why Max 20x Started Feeling Scarier

Claude Opus 4.7 changed its tokenizer, but available samples do not suggest Japanese prose itself is the main reason Max 20x feels tighter. The bigger issue is Claude Code context: English tool schemas, code, logs, JSON, diffs, and parallel agent sessions.

GPT-5.5 Prompting Guide: How to Write Outcome-First Prompts for Agents, Tools, and Structured Outputs

A practical GPT-5.5 prompting guide for developers: outcome-first prompts, reasoning effort, tool workflows, structured outputs, prompt caching, and migration examples.

How to Use Codex in 2026: A Practical Guide to the App, CLI, Skills, and Automations

How to use Codex in 2026: a practical guide to the app, CLI, prompts, skills, AGENTS.md, and automations so it becomes part of your real daily workflow.

Claude Code vs Cursor in 2026: I Used Both Daily for 6 Months. Here’s What Actually Ships Faster.

After six months of daily use shipping real projects from rural Japan, the "which is better" question misses the point. Claude Code wins on autonomy and refactoring at scale; Cursor wins on flow-state editing and model flexibility. A practical split — and the $40/month workflow that beats picking one.