Lesson 8: Self-Scheduling — A Tour of Agents

ChatGPT deep research spawns sub-tasks autonomously. Build a self-scheduling agent with a task queue and budget — what CrewAI task delegation does.

Self-Scheduling: The Agent Decides What's Next You've seen this in Claude: ask it to "refactor this module" and it decides on its own to read related files, check tests, update imports — tasks you never explicitly asked for. Or ChatGPT's deep research mode: you ask one question and it spawns multiple research threads. So far, you decide what the agent works on. A truly agentic system decides for itself. The trick: schedule_followup is just a tool. The LLM calls it like add. The side effect: a new task enters the queue. The outer loop processes tasks until the queue drains — or the budget runs out. > Framework parallel: CrewAI calls this "delegation." AutoGen calls it "nested chats." The budget cap (max_tasks) is the difference between a useful agent and a billing incident.

Concepts covered

self-scheduling
task queue
BFS
convergence
budget

Framework comparison: CrewAI task delegation, AutoGen nested chats — BFS over a dynamic work queue.

All lessons in this course

8 / 9