#coding-agents

Wiki 7

  • Open and closed models are on different exponentials Nathan Lambert on coding agents proving users pay a premium for top closed models, while open models take the larger, slower diffusion market
  • Tokenmaxxing NYT-coined term for the status game of consuming the most AI tokens; consumption metrics divorced from output, sometimes formalized in internal leaderboards
  • Typed De Bruijn indices for LLMs Replace variable names with typed slot references (`@Type.n`) so naming-related LLM error modes — misleading names, reused names, lost references — become impossible
  • What comes next with open models Nathan Lambert on why open models should stop chasing the closed frontier and build small, specialized models that closed agents call as tools
  • Which Tools Do Claude Code, Codex and Cursor Choose? Armature's 16,893-session study of which third-party services coding agents pick, run by a firm that sells vendors a better pick rate
  • Your CEO is suffering from AI psychosis HandyAI's case that executive AI enthusiasm has decoupled from output — Garry Tan's gstack, "tokenmaxxing" leaderboards, sycophancy research, and platforms that simulate management
  • Your Clippy config should be stricter Selective Clippy lints to recover the "if it compiles, it works" feeling — especially relevant when coding agents are writing the patches

Toolbox 3

  • agent-manager tmux TUI listing every coding-agent session with live status, quick prompts, worktrees, and line-comment diff review sent back to the agent
  • Fresh Terminal editor with VS Code-style keys, mouse, LSP and QuickJS plugins, plus a worktree orchestrator for coding agents
  • showagent TUI listing coding-agent sessions from every CLI, with resume, branch and cross-agent conversion