Working with Multiple Agents in Parallel

One of the biggest productivity multipliers with the Cosine CLI is learning to run multiple agents at the same time. Because agents work asynchronously in the background, you don’t need to watch them — you can fire off several tasks at once and check back when they’re done.

Why Run Multiple Agents?

Agents can take time, especially for complex research or multi-step tasks. Instead of sitting idle while waiting for one to finish, you can:

• Fire off a second agent in the same folder with a different task
• Open an agent in a different folder for a completely separate project
• Use different models in different windows to get varied perspectives on the same problem

This is the practical shift from “using AI” to multiplying your productivity — you become the one reviewing and directing, while agents do the work in parallel.

Running Multiple Agents in the Same Folder

You can open several CLI windows pointing to the same folder. Each window is its own agent session with its own context, but they all share the same files on disk.

Example workflow:

• Window 1: “Research the last 20 AWS blog posts from competitors and save a summary.”
• Window 2: “Look at the research files in this folder and draft the next 10 blog post ideas.”

The second agent can read the files the first one is writing — so they naturally hand off to each other without you needing to coordinate manually.

Running Agents in Different Folders

Each folder is its own agent with its own agent.md and context. You might have:

• ~/blog-posts/ — a content agent writing blog posts
• ~/sales-outreach/ — a Harvey Specter–style sales agent drafting emails
• ~/call-summaries/ — a summarisation agent processing call recordings

All three can run simultaneously. You jump between them, reviewing completed work and firing off the next request each time you return.

The Right Workflow Loop

1. Open an agent and give it a clear task.
2. Switch to Auto mode (Shift+Tab) so it works without interruption.
3. Open another window or switch to another folder.
4. Give that agent its task.
5. Come back to the first agent when you’ve finished with the second.

By the time you cycle back, the first agent has usually made significant progress — or is done entirely.

How Many Agents Can You Run?

There’s no hard limit. In practice, 2–4 active agents at any time is manageable for most workflows. Some power users run 8 or more simultaneously across different projects.

The limiting factor is usually your own ability to review and direct their output rather than any technical constraint.

Key Takeaways

• You don’t need to watch an agent work — start it, switch away, come back when it’s done.
• Multiple CLI windows can run in the same folder or different folders simultaneously.
• Agents in the same folder can share context by writing and reading files on disk.
• The most productive workflow is treating yourself as the director — reviewing and redirecting multiple agents rather than watching any one of them.

---

Next: Plan Mode: Think Before You Act