Skip to content
sec/atlas
back to the map
concept
stage: choose
2 min read

When to use what

Start from the task and its risk, not from the tool you like. A small decision table covers most days.

Claude CodeOpenCodeCursorherdrAiderOh My OpenAgent (OMO)

Start with the task

  • A quick question about code, or a small edit in a file you have open: IDE chat or inline edit (Cursor, Copilot) or a terminal agent in .
  • A feature across several files in a repository with tests: a terminal agent (Claude Code, OpenCode, , Codex CLI) with approvals on for writes and shell.
  • A large mechanical change (rename, migration, upgrade): an agent with strong edit tools and a clear rule, run in a branch with the test suite as judge.
  • Many independent tasks: several agents in separate git worktrees, organized with a multiplexer such as herdr, or an orchestration layer such as on OpenCode.
  • A throwaway prototype or script: full autonomy is fine. Run it in a sandbox or a scratch directory.
  • Security-sensitive code (authentication, payments, crypto, parsing): use the agent for a draft and for tests, then review line by line with a person who knows the area.
  • Unfamiliar codebase: ask the agent to map and explain first, read-only, before any change.

Match autonomy to risk

Think of two dials. Autonomy is how much the agent does without asking. Risk is the damage a wrong change can cause. Keep the product of the two small: high autonomy only where mistakes are cheap and reversible (a branch, a sandbox, a script), and approvals where they are not (production config, credentials, data migrations, deletions).

Choosing the harness

  • Want one polished, vendor-supported experience: Claude Code (Claude models) or Codex CLI (OpenAI models).
  • Want to swap models and providers freely, or run local models: OpenCode or OMP.
  • Want built-in roles and autonomous multi-agent work on OpenCode: add OMO.
  • Run several sessions at once and lose track of them: herdr or tmux.
  • Prefer a graphical IDE: Cursor or an IDE extension.
  • Want tiny, git-first, predictable commits: Aider.

Signs you chose wrong

  • You keep accepting changes you did not read.
  • The agent loops, repeating the same failed fix: shrink the task or change the model.
  • Costs surprise you: set limits, use cheaper models for easy steps, keep context small.
  • You cannot explain what the code does: stop and read it before building more on top.
  • Each new tool layer adds another thing to configure and trust: remove what you do not use.
Plan modeA read-only mode where the agent explores and proposes a plan for approval before it edits anything.see in glossary →OMPoh-my-piAn open-source terminal coding agent that began as a fork of the minimal Pi agent and added structured edits, language-server and debugger operations, browser access, plugins and subagents.see in glossary →OMOOh My OpenAgent · oh-my-opencodeA plugin for OpenCode that adds a team of specialized agents and an autonomous multi-agent mode. It extends a harness; it is not one.see in glossary →

connections on the map

updated on 14/10/2026 · sources to go deeper · report an error

Study before

test what you learned

Loading questions…