Thread: Orchestrating Claude Code Agents

1/ I've been running 4 Claude Code agents at once across different terminals.

My laptop is melting.

A few observations 🧵

2/ Claude Code has 3 modes:

You're basically adjusting how much autonomy to give it.

3/ Different tasks need different modes.

Risky refactor? Plan mode. Boilerplate? Auto-accept. Exploring a problem? Chat.

So I started running multiple agents for different jobs. That's when things got interesting.

4/ The problem: my M2 Max chokes at 4+ agents.

Each Claude Code instance spawns multiple node processes. Heavy API calls. Memory spikes.

I've crashed VS Code more times this week than all of last year.

5/ Solutions I'm exploring:

6/ The question I keep coming back to:

How do you architect a workflow when you have multiple agents that can all work at the same time?

This isn't a Claude Code problem. It's a genuinely new problem.

7/ What I actually want is something that sits above the agents.

Shows me what each one is doing. Knows not to run 4 heavy tasks at once. Stops two agents from editing the same file.

Basically: air traffic control for AI agents.

8/ Anyone solved this?

Running multiple agents locally? Codespaces? Something else?

Genuinely asking. DMs open.

Alternative Hooks

Option A (problem-first): "My laptop crashed 3 times today running Claude Code.

Not because the tool is broken—because I was running 4 agents at once.

Here's the multi-agent orchestration problem nobody's talking about 🧵"

Option B (question-first): "How do you run multiple AI coding agents in parallel without your machine catching fire?

Asking for a friend (the friend is me) 🧵"

Option C (spicy take): "Claude Code's real limitation isn't context length.

It's that there's no orchestration layer for running multiple agents.

Thread on what I think is missing 🧵"