Though parallel agents already exists, we built this out differently in Pochi. Most other tools operate inside a single editor tab. That means there is at most one conversational state to interact with.

Parallel Agents in Pochi use separate Git worktrees, so each task has its own working directory, branch, chat history, and terminal environment. Task state stays isolated.

From a UX standpoint, the important bit is how it surfaces in the editor: each agent is a separate tab, each tab bound to its own worktree. You can diff, commit, discard, or merge worktrees independently. You can run two model-generated solutions side-by-side and compare outcomes without the branches stepping on each other.

Under the hood it’s git worktrees with orchestration to bind each worktree to its own agent state. If you're familiar with git worktree add you’ll recognize what’s happening. We expose the worktrees directly through VS Code (Source Control + Pochi tabs), so there’s no manual git worktree management - each agent just is a worktree with its own runtime context. Goes to show how some simple changes dramatically changes the UX.