Multi-agent coding systems are *crazy* good. Feels like cheating. 2-3x better than single-agent. Tutorial + prompts + code coming soon. Watch 2min sneak peak.

And the entire point of a system like this is that it’s just a handful of text files. No app, no subscription, no anything. Just a few core markdown files and prompts. Extremely customizable. Simple is better with this stuff.

Simple multi-agent systems and simple skill libraries are the 2 highest-impact things you can add to your workflows right now. I should probably do a video on skill libraries as well. Think this but for coding:

Excited to see what you've put together. Crazy good is putting it lightly. If it can check each other's work and instantly communicate results between agents, that's a capability you just don't get with prompting alone..will it do that?

It’s honestly pretty simple. The highest leverage point for it going well is the plan approval step. Way way way too many people just auto accept plans and wonder why they get bad results. Communication layer is literally just a markdown file with a prompt paired with it.

A bit clickbaity. Would expect multi-agents systems to be at least 2x a single agent by default. What's the euphoria for? What am I missing here?

The fun! You’re missing the fun! We’ve got intelligence on machines that we can harness together in swarms to boost productivity! We live in the FUTURE

Approaching Opus usage limit

Let me sell the other kidney to feed 4 Claude Code accounts then

If you need/want your agents to work entirely in parallel we're exploring this idea with an OSS project which provides environments and sandboxing for agents.

You know you can run multiple agents within one Claude code instance. It can split off into 4 agents.

The problem with subagents in a single core agent is you need: - separate context windows - full observability - ability to interrupt each one