Claude Code Review Needs a Second Model
Claude Code review measured on 62 of our own pull requests. What it catches, what it waved through, the local command against the GitHub App, and the second model.
Read article →FIELD NOTES · BUILT FROM REAL RUNS
Systems for developers
who run agents.
Technical guides, honest benchmarks and operating notes from building a native cockpit for Claude Code, Codex, Gemini and OpenCode.
LIBRARY · 13
Read the field notes.
Practical writing for solo developers scaling their output with coding agents.
OpenCode vs Claude Code, Stop Choosing
OpenCode vs Claude Code, measured not argued: what the free alternative really costs since Anthropic closed subscription OAuth, and why we run both instead of choosing.
Read article →Codex Pricing Is Not the $20 You Think
Codex pricing decoded: the plan grid, what a credit really costs, how to read your usage in the CLI, and what 30 days of Codex on a real repo billed us per shipped change.
Read article →Claude Code Agent Teams Are Not Pointless
Claude Code agent teams, measured. The env var that turns them on, how the lead and mailboxes work, what five teammates cost in RAM and tokens, and when subagents win.
Read article →Six Claude Code Worktrees, Only Four Paid Off
Claude Code worktrees end to end: the native --worktree flag, .worktreeinclude, what six parallel sessions cost in RAM and quota, and how to spot the stuck one.
Read article →Codex vs Cursor, We Ran the Same Ticket
Codex and Cursor compared on the same real ticket: wall clock, human interventions, cost per feature shipped, and what breaks after twenty minutes.
Read article →The 9 Codex Skills That Earn Their Context
Codex skills explained: the paths that actually work, the 2% metadata budget, a real SKILL.md, and nine skills worth installing.
Read article →Cursor vs Claude Code, Stop Comparing Benchmarks
Cursor and Claude Code compared on token cost, context, pricing and the terminal harness nobody measures. The routing rules we use instead of leaderboard scores.
Read article →Best Vibe Coding Tools That Ship Real Apps
The vibe coding tools that survive contact with production, split by browser builders versus terminal agents, with real pricing, security data and the handoff nobody plans for.
Read article →The Codex CLI Guide OpenAI Never Wrote
Install Codex CLI, pick between Sol, Terra and Luna, understand the real 272K context cap, wire AGENTS.md and config.toml, and know when the app beats the terminal.
Read article →The Best Terminal App for Mac, Benchmarked
AITerm, Ghostty, iTerm2, Warp, Kitty and Alacritty benchmarked on latency, memory and panes. AITerm wins for piloting AI coding agents, Ghostty for plain shell work.
Read article →Codex vs Claude Code, We Run Both Daily
Codex CLI and Claude Code compared on benchmarks, real cost per task, context, config files and safety. The routing rules we use to send each task to the right agent.
Read article →How to Use Claude Code Like a Power User
Install Claude Code, keep CLAUDE.md under 200 lines, pick between Sonnet 5 and Opus 5, and stop burning context. The setup we run daily, with real numbers.
Read article →FROM THEORY TO SHIPPED CODE
Put the system
to work.
Run 20+ coding agents and see exactly where each one is. The full product is free for seven days.
Download for macOS 45 MB · no account · no card