What is Supervised Execution?

Supervised Execution: The preferred mode of operation for Clean Vibe developers. Unlike "Background Autonomy," where an agent works unsupervised, Supervised Execution keeps the human "In-the-Loop" at every decision point, ensuring that no code is written that isn't understood and approved by an engineer.

The Supervised Workflow

  1. The Intent: Human provides a Clean Prompt.
  2. The Plan: Agent analyzes context and proposes a plan.
  3. The Checkpoint: Human reviews and adjusts the plan.
  4. The Implementation: Agent writes code for one atomic part of the plan.
  5. The Verification: Human reviews the diff, tests the logic, and "signs off."
  6. Repeat.

Why It's Faster (Long-Term)

Supervised execution feels slower than letting an agent run overnight. However, it eliminates the 3-day "Digital Archaeology" sessions needed to fix unsupervised disasters. It maintains Sustainable Velocity by preventing the accumulation of Synthetic Debt.

Statistics

1/4
debt accumulated vs unsupervised
Source: Ch 14
High
Level of Developer Satisfaction
Source: Research

The "Agent-as-Intern" Model

Supervised Execution treats an AI agent like a highly capable, high-speed junior intern. You give them specific tasks, check their work frequently, and never give them the keys to the production database unsupervised.

Related Terms

Book Reference

  • Chapter 14: Clean Agents — how to configure and use supervised execution.
  • Chapter 18: The Clean Vibe Manifesto — "We refuse to let automation run unsupervised."

Collaborate, don't abdicate