Agent Task Flowchart 🍓🔀

When to send agents vs. when to babysit them

Agent Usage Patterns

🍓

Simple Self-Contained Tasks

Send it and forget it - perfect for focused, single-purpose work

→Build calculator app

→Create conversion tool

→Generate widget

🔍➡️⚙️

Multi-Stage Agent Pipeline

Send agent to research, review results, send another to implement

→Market analysis → feature build

→Data gathering → visualization

→Requirements → architecture

🧑‍✈️

Hands-On Collaboration

Babysit the agent - surgical edits, constant supervision

→Refactoring existing codebase

→Debugging complex issues

→Iterative design changes

Interactive Decision Tree

Is the task self-contained?

Agent Personality Matrix

🐉 Here Be Dragons

🔍⚙️ Background Agents

🧑‍✈️ Copilot Zone

🍓 Sweet Spot

← Low Autonomy | High Autonomy →

← Simple | Complex →

🎯

🧑‍✈️ Copilot Zone

Hands-on collaboration for precise edits

• Refactor existing code

• Debug complex issue

• Surgical improvements

💡 Practical Tips & Wisdom

Signs you're babysitting too hard:

You're reviewing every line, making constant corrections, and feeling exhausted. Time to switch to copilot mode or break the task down.

When background agents actually work:

Clear requirements, self-contained scope, well-defined success criteria. If you can't verify success automatically, it's not ready for background mode.

The "commit immediately" test:

Would you commit what the agent gave you without reading it? If no, you're in the wrong workflow. Either scope it better or use copilot mode.