{"slug": "agent-manager-run-multiple-agents-without-the-chaos", "title": "Agent Manager: Run Multiple Agents Without the Chaos", "summary": "Kilo Code released Agent Manager, a full-screen workspace that runs multiple AI agents in parallel within isolated git worktrees, preventing file overwrites and merge conflicts. The tool, accessible from the sidebar, uses a kanban-style board to track tasks and allows users to compare different AI models on the same problem before committing changes.", "body_md": "# Agent Manager: Run Multiple Agents Without the Chaos\n\n### Git worktrees, model comparison, and zero merge conflicts — without leaving VS Code\n\nI’ve got six things I want an AI agent to work on right now. Refactor auth. Update docs. Add Stripe billing. Write tests. Migrate the database schema. Fix that flaky CI pipeline.\n\nThe obvious move is to open six tabs and let them rip. But what actually happens is agents overwriting each other’s files, merge conflicts piling up, and context bleeding between tasks. I end up spending more time cleaning up the mess than shipping code.\n\nAgent Manager is designed to prevent that kind of overlap by isolating each task in its own worktree.\n\n## What It Is\n\nThe Agent Manager is a full-screen workspace inside Kilo Code: for running multiple agents in parallel. Each agent gets its own git worktree — a completely isolated copy of your repo with its own branch. The agent working on “Refactor auth” doesn’t share a working directory with “Update docs.” They’re in separate directories on disk.\n\nYou open it with the multi-agent button in Kilo’s sidebar. The layout is a kanban-style board: To Do, Doing, Done. Each card represents one agent running in one worktree.\n\n## How It Works\n\nClick “New Worktree.” Write your prompt — something like “Implement Stripe checkout with webhook verification.” Pick your model and mode. Hit Create.\n\nKilo spins up the worktree, initializes the agent, and starts working. The card shows live status: reading the codebase, planning, writing files. You can watch it happen or go do something else.\n\nThe agents run in parallel. I regularly have four or five going at once across different tasks. None of them can interfere with each other because they’re physically in separate directories with separate branches.\n\n## Compare Models on the Same Task\n\nYou can also create multiple worktrees with the same prompt but different models. Claude Sonnet for one, GPT 5.5 for another, Claude Opus for a third.\n\nThey work the same problem simultaneously. When they finish, you review both diffs and pick the approach you like better. One might have cleaner architecture. Another might have better error handling. This gives you a practical way to compare outputs before choosing which implementation to keep.\n\n## Review and Ship\n\nWhen an agent finishes, click “Review Changes” on the card. You get a full diff — syntax highlighted, side by side. You can see exactly what files changed and what the agent did before you commit anything.\n\nWhen you’re satisfied, click “Open in VS Code.” It opens that worktree as a full folder in a new window. IntelliSense, debugger, terminal, test runner all work because it’s just a normal VS Code window. When you need to tweak something, run tests, or debug, you’re already in the editor. No switching to a separate app.\n\n## Built-in Terminal\n\nAgents will sometimes hit errors during implementation. When one does, you click the card and open the built-in terminal. It’s already cd’d into the correct worktree. You don’t need to figure out which directory you’re in or switch to a separate terminal app.\n\nFix the issue, and the agent picks up where it left off.\n\n## Why This Matters\n\nBefore this, running multiple agents meant accepting that they’d conflict. You’d serialize your work — finish one task, then start the next — which reduces the benefit of parallel agent work.\n\nThe Agent Manager gives you isolation by default. Separate worktrees mean separate file systems. Each agent only sees its own branch. You can track each task independently without manually coordinating directories or branches, and you never leave VS Code to manage any of it.\n\nWhen you’re done, drag the card to “Done” and merge the branch like any other feature. Standard git workflow.\n\n## Try It\n\nThe Agent Manager is available now in Kilo Code. Open it from the multi-agent button in the sidebar, create a couple worktrees, and let them run in parallel. It’s useful when you want parallel agent work without overlapping file changes.\n\nOne quick compatibility note: v5 of the VS Code extension will no longer be supported after July 31st, so make sure you’re on the latest version to keep getting updates and Agent Manager improvements.", "url": "https://wpnews.pro/news/agent-manager-run-multiple-agents-without-the-chaos", "canonical_source": "https://blog.kilo.ai/p/agent-manager-run-multiple-agents", "published_at": "2026-06-08 20:31:07+00:00", "updated_at": "2026-06-11 18:45:08.847622+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "ai-products"], "entities": ["Kilo Code", "Agent Manager", "Stripe"], "alternates": {"html": "https://wpnews.pro/news/agent-manager-run-multiple-agents-without-the-chaos", "markdown": "https://wpnews.pro/news/agent-manager-run-multiple-agents-without-the-chaos.md", "text": "https://wpnews.pro/news/agent-manager-run-multiple-agents-without-the-chaos.txt", "jsonld": "https://wpnews.pro/news/agent-manager-run-multiple-agents-without-the-chaos.jsonld"}}