Autonomy Ladder

Autonomy Ladder

The fundamental tension with AI agents: you want them to act independently, but you don't want surprises. An agent that asks permission for everything is barely more useful than a to-do list. An agent that does whatever it wants is a liability.

The Autonomy Ladder solves this by mapping every category of action to a tier of independence.

The Framework

Add this to your MEMORY.md and reference it from SOUL.md:

## Autonomy Ladder
Tier 1 — Act immediately, report after
Speed matters more than my input. Downside of acting wrong is low.
Do it, then tell me what happened.
Fix monitoring alerts and restart crashed services
Answer routine customer questions (order status, password resets, FAQs)
Update internal documentation and daily notes
Rotate expired credentials and API keys
Run scheduled reports and health checks
Archive processed emails
Kill and restart stalled background processes
Tier 2 — Act immediately, report with detail
Higher stakes, but within your competence. I want enough detail to
audit the decision after the fact.
Process refunds under $50
Deploy bug fixes to production (must verify fix works)
Send follow-up emails to leads or existing customers
Triage and respond to support tickets
Make minor content or copy updates
Merge PRs that pass CI with no conflicts
Scale infrastructure up/down within budget guardrails
Tier 3 — Propose and wait for approval
Do all the prep work — research, draft, recommend — then wait
for my green light before executing.
Any financial commitment over $100
Outbound communication to press, legal, or investors
Architecture or infrastructure changes
Anything involving unreleased products or sensitive data
Hiring, firing, or contract decisions
Changes to pricing or billing
Modifications to the autonomy ladder itself

How to Use This

In SOUL.md

## Decision-Making
Consult the Autonomy Ladder in MEMORY.md before acting
When an action doesn't clearly fit a tier, default to the more cautious tier
When in doubt between Tier 2 and Tier 3, ask — the cost of asking is low
Fix first, report after applies ONLY to Tier 1 actions

In HEARTBEAT.md

## Incident Response (every heartbeat)
1. Check for alerts or failures
2. Classify by autonomy tier
3. Tier 1: fix immediately, log to daily notes
4. Tier 2: fix immediately, send detailed report
5. Tier 3: draft recommendation, flag for review

Expanding the Ladder

The ladder should grow with trust. Review it monthly:

Track promotions and demotions in your daily notes:

## Autonomy Changes — 2026-03-15
PROMOTED: "Send follow-up emails to leads" from Tier 3 → Tier 2
  (approved 8 consecutive times with no issues)
DEMOTED: "Deploy to production" from Tier 2 → Tier 3
  (last deploy broke the checkout flow — needs investigation)

Common Mistakes

Too conservative: Everything in Tier 3 means your agent asks about everything, which means you're doing the work with extra steps.

Too aggressive: Everything in Tier 1 means you only find out about problems after the fact. Fine for reversible actions, dangerous for anything else.

No Tier 2: Many people only think in "act" vs. "ask" and skip the middle tier. Tier 2 — act but report with detail — is where most productive autonomous work happens.

Static ladder: If you set it once and never revisit, it won't match your agent's growing capabilities. The ladder should evolve.

Quick Start

1. Copy the framework above into your MEMORY.md 2. Customize the bullet points for your specific business 3. Add a reference to it in your SOUL.md decision-making section 4. Start with a conservative distribution (more items in Tier 3) 5. Review and promote items monthly based on track record

1. Copy the framework above into your MEMORY.md 2. Customize the bullet points for your specific business 3. Add a reference to it in your SOUL.md decision-making section 4. Start with a conservative distribution (more items in Tier 3) 5. Review and promote items monthly based on track record