Conductor for AI Development
Gareth Morgans /
I've started using an app called Conductor which is a Mac app that lets you spawn multiple coding agents at the same time. If anybody is looking to speed up their AI development workflow, then I would certainly recommend it. I've tried it out on a couple of projects and so far it's great. It is really speeding up my development workflow, which was quite “linear” previously (that’s not a reference to Linear which is another multi-agent tool I want to try).
I like using Cursor, that is my preferred IDE, so I was a little bit nervous to step outside of that. However, what Conductor allows you to do is avoid sitting around and waiting for the AI to finish a task. You can basically spawn multiple instances of your project in Cursor and work on several bits all at the same time. Conductor then handles the pull requests for you to merge those different bits back into the project once they are finished. Instead of having a one-track development workflow, you are able to work on multiple features or fixes simultaneously, merging them back into the project when complete.
It can feel like whacking the accelerator down on your car and letting go of the wheel at times because you are working so much faster. There is a tendency to want to move quicker with the project as a whole, but you have to be strict with yourself. You must ensure you are checking, testing, and validating what these different things are doing before you merge them back into your main project.
How It Works
Conductor is the central dashboard where you open your connected GitHub repos as projects
Under each project, you create workspaces, each one is an isolated git worktree with its own branch, files, and terminal
Each workspace runs a Cursor Composer session, authenticated with a Cursor API key (set in Settings, Harnesses, Cursor)
Starting a new task means creating a new workspace, which opens as its own Cursor window pointed at that isolated directory, rather than one Cursor instance juggling everything
If you're also running Claude Code in parallel, those sessions authenticate through your existing Claude CLI login (Pro, Max, or API key) instead of the Cursor key
Because each workspace has its own checkout, several Cursor tasks can run at once without one overwriting another's changes
Finished work gets reviewed in Conductor's diff viewer, then merged or turned into a PR from there
Key Features & Versatility
The tool offers a great set of built-in features, especially for reviewing. In review mode, it spawns another agent, this time in Opus and then provides feedback. This has also sped up my workflow.
There is loads of versatility with it:
Parallel agents - run several Claude Code, Codex, Cursor, or OpenCode sessions at once, either across separate workspaces or multiple chats in one workspace
Diff viewer - review an agent's changes before merging, without leaving Conductor
Scripts - setup scripts and run scripts to get a workspace ready or start dev servers automatically
Testing - spotlight testing to check agent output before you merge
Todos - track what each agent is working on and what needs your attention
MCP support - carry over your MCP servers so agents have the same tools and context
Slash commands - quick actions inside a workspace
Checkpoints - save and roll back to earlier states in a session
Deep links - jump straight to a specific workspace or task
PRs - link a GitHub or Linear issue to a workspace and go straight from task to reviewed pull request
Conductor Cloud - run agents on remote microVMs instead of your Mac, so they can keep working for hours without your machine staying on
I am on the free plan at the moment, but I think I will be looking at how we can use this across our team and probably move on to the paid plan as soon as I can. I am quite interested in the cloud version - you can work remotely without having to run it on your local machine, which opens up loads of possibilities. However, if you’re a single dev then you’ll be able to go very far on the free plan and may never need to upgrade.