AI Tools
Local LLMs on macOS in 2026: when they're worth the GPU
Local LLMs got dramatically better in 2025-2026. They're competitive with frontier APIs for some workflows; not all. Here's the honest picture.
Honam Kang6 min
AI Tools
AI Tools
Practical guides for Claude Code, Cursor, Codex, cmux, and the rest of the AI dev stack.
AI Tools
Claude Code memory without polluting global config
Claude Code's memory feature is powerful but easy to misuse. The pattern that scales — what to put in global memory, what to put per-project, what to never persist.
Honam Kang6 min
AI Tools
File-context strategies for AI agents: what to feed, what to skip, what to summarize
When an AI agent has access to your whole repo, it doesn't read your whole repo. Here's how to choose what enters context, what stays out, and how that decision affects output quality.
Honam Kang4 min
AI Tools
When the agent is wrong: a debugging protocol
AI agents fail in specific ways. The right debugging response depends on the failure mode. Here's the structured protocol that catches drift fast.
Honam Kang6 min
AI Tools
cmux multi-session setup: running 5 AI agents in parallel without losing your mind
cmux turns one terminal window into a tile of independent agent sessions. Here's how to wire it up, name your sessions, and integrate it with a file manager so context never gets lost.
Honam Kang3 min
AI Tools
Context engineering: the hidden lever for agent quality
Prompt engineering peaked. Context engineering — what's in the agent's working memory, what's not — is the 2026 leverage point. Here's the practical playbook.
Honam Kang7 min
AI Tools
Claude Code vs Cursor: how each handles your file system (a 2026 comparison)
Both tools edit files. They have very different opinions about which files, how many at once, and where their working memory lives. Here's the honest comparison from someone who runs both daily.
Honam Kang5 min
AI Tools
Codex CLI: the underrated agent for terminal-heavy work
OpenAI's Codex CLI doesn't get the attention Claude Code or Cursor do, but it's surprisingly capable for terminal-native workflows. The honest review.
Honam Kang6 min
AI Tools
Prompt caching for cost: cutting Anthropic bills 60%+
Prompt caching is the highest-ROI optimization for heavy Claude API users. Done right, it cuts costs 50-80%. Here's how — with concrete patterns.
Honam Kang5 min
AI Tools
Custom Claude skills: when to write one (and when not to)
Claude skills are reusable agent capabilities. They're powerful — but writing one for the wrong workflow is wasted effort. Here's the practical guide.
Honam Kang5 min
AI Tools
Cursor Composer vs Agent mode: when to use which
Cursor's Composer and Agent mode look similar but optimize for different work. Composer is for in-flow edits; Agent is for delegated multi-step. The decision tree.
Honam Kang6 min
AI Tools
Aider vs Claude Code for refactoring: a side-by-side
Aider and Claude Code both do agent-style coding from a terminal. They diverge on git workflow, model flexibility, and edit precision. The honest comparison for refactor work.
Honam Kang6 min