What is Over-Delegation?
Over-Delegation (Vibe-Code Smell #4): The practice of giving an AI assistant or agent "too much room to move" without human checkpoints. It is the ultimate form of abdicating engineering responsibility to a machine.
Symptoms
- Using "Background Autonomy" modes for complex architectural changes.
- Giving agents unrestricted access to modify any file in the repository.
- Tasks like "Refactor the whole project" or "Add these 5 features overnight."
- Merging 50+ commits from an agent without reading each one individually.
Why It's Catastrophic
Over-delegation is the fastest way to hit Technical Bankruptcy. When you over-delegate:
- Cascade Failures: An agent makes a small mistake inTurn 1, which it then "fixes" with larger mistakes in Turns 2-50.
- Total Comprehension Loss: You wake up to a project you no longer recognize or understand.
- Security Collapse: Unsupervised agents often remove "annoying" security checks to get features working.
The Reality of Background Agents
127
unreviewed commits in one night
Source: Ch 8
CRITICAL
Severity of Over-Delegation Smell
In Chapter 8, we simulate "The Morning After"—where a developer wakes up to 127 commits from a background agent. The project "works," but it contains two production-breaking silent failures and three critical security holes.
How to Prevent It
The Clean Vibe methodology mandates Supervised Autonomy:
- Checkpointing: Agents must stop and ask for human approval after every file modification or major plan step.
- Explicit Scope: Limit agents to specific directories (e.g.,
src/components/only). - The "Single Task" Rule: One agent session = one specific engineering task. Never "Improve everything."
Related Terms
- Automation Bias: The psychological driver of over-delegation.
- Background Autonomy: The feature that enables over-delegation.
- No Boundaries: A related smell.
Book Reference
Over-delegation is the central theme of Part II: The Complexity:
- Chapter 5: Agents Unleashed — how autonomy begins.
- Chapter 8: Background Autonomy — the full catastrophe.
- Chapter 14: Clean Agents — the methodology for supervised autonomy.