This week, I’m sharing something more practical!
I’ve been a fan of Claude code sub-agents for a while. I mostly used them to split up context-heavy processes into smaller, focused agents.
Recently, while chatting with my colleague Brad Carter, he mentioned the idea of replicating the different personas inside a startup — having different AI roles work together. It sounded odd at first, but I had to try it for myself.
After experimenting with a few personas, one combination stood out: a software engineer and a code reviewer.
I think the magic comes from how the two sub-agents collaborate — the reviewer always starts with a fresh perspective. This helps catch any hallucinations or deviations from the plan early on, preventing them from spreading through an entire coding session.
Setting Up Your Agents
I’ve published the agents in my Claude marketplace for easy access. You can also copy-paste them freely.
claude plugin marketplace add https://github.com/SBoudrias/claude-marketplace
Then open Claude, run /plugin, and select Browse and install plugins. Choose the marketplace, then install startup-subagents.
How to Use
I’ve included a helper command, but it’s really just a one-liner prompt:
Orchestrate work between the senior-software-engineer and the code-reviewer to implement [...fill in the blank!]
In practice, I often ask Claude to help me draft a technical plan inside a file like PLAN.md, then run:
/orchestrate @PLAN.md
For smaller tasks, I just run something inline, like:
/orchestrate fix failing CI on the PR
Conclusion
That’s it — simple and powerful. Give it a shot and let me know how it goes!
If you’re still struggling to get high-quality code from AI agents, check out my previous post. The most common mistake I see is delegating the thinking. Once you describe how the code should be implemented, you’ll realize the AI is there to assist — but you’re still in the driver’s seat.
Top comments (0)