← Deep Catalogue  ·  F-01 Foundational
AUT-DSN-01Tier IBounded Autonomy

Autonomy Tier Classification

Classify every agent by the reversibility and blast radius of the actions it may take, and bound its permissions accordingly.

Runtime: Reason / PlanAgentOps: Design & ScopeSee on the matrix →
AUTONOMY TIER LADDER T4 · Fully Autonomousagent → action → outcome T3 · Bounded Autonomouspre-declared envelope; post-hoc review T2 · Assistiveagent proposes · human approves each material step T1 · Advisoryagent suggests · human actsEach tier prescribes HITL density, tool envelope, review cadence, blast radius.
In Plain English

This page is the complete instruction page for one pattern called "Autonomy Tier Classification." It explains the problem this pattern solves, the idea behind the solution, when you should (and should not) use it, and what happens afterward — both the good effects and the costs.

This matters because building AI agents is not just about making them clever. It is also about making them safe and predictable. Following a well-tested pattern like this one helps avoid common mistakes, and shows you exactly which safety rules and regulations it connects to, listed under "Standards Mesh" on this page.

Context

Organisations deploy agents of vastly different risk. An agent that drafts an email needs different governance than one that moves money or updates production data. Without explicit tiering, every agent defaults to the permissions of its most capable component.

Problem

Treating all agents identically either over-constrains the harmless ones or under-governs the dangerous ones. Regulators now require a documented basis for autonomy decisions.

Forces

  • Productivity vs safety
  • Flexibility vs auditability
  • Developer velocity vs organisational risk

Solution

Define a small, fixed ladder of autonomy tiers (e.g. T1 advisory · T2 assistive · T3 bounded-autonomous · T4 fully autonomous). Each tier prescribes: maximum blast radius, mandatory HITL checkpoints, tool-access envelope, and review cadence. Classify each agent at design time; re-classify on scope change.

Applicability

  • All agent deployments under IMDA MGF
  • High-risk use cases under EU AI Act Art. 6–7
  • Any agent touching regulated data or transactions

Anti-Patterns

  • "Autonomy" undocumented and left to the LLM
  • Production agents escalated from prototypes without re-tier

Consequences

  • +Explicit risk posture for every agent
  • +Auditable basis for HITL and guardrail density
  • −Requires classification discipline
  • −Tier creep if not reviewed