Skip to content

Lift 4: The Autonomy Slider

Where We're Starting

You've automated a lot. Eval failures become work items. Agents pick them up in isolated worktrees. Quality gates block deployment when evals regress. The system is increasingly capable of operating without your direct supervision.

But some decisions are happening without you that you're not comfortable with. The alert engine sent a notification based on its own analysis — was that the right call? An agent refactored the analysis engine and the evals pass — but the approach is completely different from what you would have chosen.

If you step in for everything, you've defeated the purpose of the system you built. If you don't step in for anything, you've abdicated responsibility for decisions that matter.

This lift gives you the frameworks to draw that line deliberately — not by default, not by accident, but by design.

What You'll Learn

  • Why AI defaults to agreement — the obedience problem — and why it matters for autonomous systems that need to flag when something is wrong
  • The autonomy slider: from "approve every action" to "intervene only when it matters" — and the reversibility principle that determines where you set it
  • Orchestration patterns that use structured disagreement to improve critical decisions
  • How to draw the deliberate line between what runs autonomously and what requires human judgment

Sections

  1. The Obedience Problem — AI defaults to agreement — and why that matters when you're trusting it with decisions
  2. The Autonomy Slider — From approve-everything to intervene-when-it-matters — and the reversibility principle
  3. The Deliberate Line — Orchestration patterns, the approval spectrum, and drawing the line by design

By the End of This Lift

  • You can describe the obedience problem and why it creates risk in autonomous systems
  • You can apply the reversibility principle to determine where human approval is needed
  • You understand the autonomy slider and can place different system components at different positions on it
  • You know at least two orchestration patterns for critical decisions: team lead + specialists, and debate/consensus
  • Your team has deliberately chosen where humans stay in the loop and where the system runs on its own
  • You can trace the full journey: doer → delegator → director — and articulate what changed at each stage