In partnership with

Smart dictation that understands you

Typeless turns your raw, unfiltered voice into beautifully polished writing - in real time.

It works like magic, feels like cheating, and allows your thoughts to flow more freely than ever before.

With Typeless, you become more creative. More inspired. And more in-tune with your own ideas.

Your voice is your strength. Typeless turns it into a superpower.

Download for Mac today

We’ve already covered what Claude sub-agents are and how they work. If you missed that overview, you can find it here.

Today, let’s take it a step further and focus on using these sub-agents effectively with Delegation and Orchestration

Delegation

Claude’s sub-agents can transform how you work with code, but most people don’t set them up in a way that makes them actually useful. The main agent won’t delegate unless the sub-agents are clear, specific, and easy to call. You need the right agent setup and task routing for this.

Agent setup

Sub-agents act like micro-services inside your repo. They live in .claude/agents/ as simple Markdown or YAML files.

Each file needs four parts:

• name in short kebab case, like bug-fixer or test-writer

• description that signals when Claude should route tasks to it

• tools kept to the bare minimum

• model (default sonnet, switch to opus only for heavy jobs)

Keeping them in the repo makes collaboration easier. Everyone has the same set, the agents are versioned with the code, and Claude discovers them without extra steps.

Task routing

Descriptions control delegation. If they’re vague, the main agent keeps the work for itself.

Direct phrasing works best. Use strong cues like “MUST BE USED” or “ONLY FOR” and tie the agent to a file path or task.

Weak: “Helps with bug fixes”
Stronger: “MUST BE USED when a .diff patch is required; ONLY outputs patch.diff; never edits source directly”

Practical rules

• Anchor to a specific folder or file type

• Use action verbs: write, patch, scan, summarize

• Keep the scope small so the agent doesn’t wander off

Orchestrating in practice

Remember progress.md? Many of my readers have reached out, saying it has significantly changed the way they use AI agents. You can read about it here.

When you have multiple agents, they won’t coordinate unless you give them a shared log. progress.md works as that link. Place it in the repo root.

# Example progress log

## 2025-09-21-test-writer
- Added failing tests for auth refresh bug  
- Artifacts: tests/authrefresh.spec.ts  
- Next: bug-fixer to create patch.diff  

## 2025-09-21-bug-fixer
- Generated patch.diff to fix refresh issue  
- Artifacts: patch.diff  
- Next: run test-runner to confirm  

Each agent appends what it did, what file was created, and which agent should run next. This gives main CC agent context to orchestrate tasks without guessing.

Sub-agents directory

I've discovered some outstanding git repos that feature practical, high-impact agents for developers. Sharing a curated selection below

Once agents are in place, be direct with instructions on the main CC agent.

Takeaway

Sub-agents let you break down your workflow into reliable steps. Each agent handles one part of the chain, the main orchestrator delegates and progress.md ties the whole flow together.

Can you take a moment to rate this edition? It helps me cater the right content to you.

👓️ Worth reading for Devs

• You should be rewriting your prompts

• Vibe coding is creating brain dead coders

📱 Stay connected: Follow me on LinkedIn and Twitter/X for daily AI tool breakdowns and quick wins