We use functional cookies to operate the site and, with your consent, Google Analytics cookies for visitor statistics. You can accept all or limit to functional only. Learn more about cookies.

← Autonomy

Autonomous Levels

The L0–L5 scale inherits the TM Forum AN baseline and describes autonomy of a single agent or product. L6 and L7 extend it for agent systems: an orchestrator that creates agents, and an ecosystem that develops its own infrastructure and rules.

Manual

System monitors only — no automated actions.

Human role: All actions performed manually.
Real-world example: A blinking server light. The admin reads logs and reacts.

Arcanada context

Live products can sit here honestly; see the list below.

Projects currently rated at this level in the registry:

Adsessor now L0 → target L4 Billing Arcana now L0 → target L3 Conversion Arcana now L0 → target L3 Overlook now L0 → target L3 Voice Agent now L0 → target L3 Argana AI now L0 → target L4 Conversation to Markdown now L0 → target L1 Orchestrator now L0 → target L3 AI Agents Catalog now L0 → target L2 Academy now L0 → target L1 Store now L0 → target L2 Community now L0 → target L1 Combateka now L0 → target L3

Assisted

Executes pre-configured deterministic subtasks (cron, pipeline step). Failures emit raw logs.

Human role: Initiates every significant action; reads logs manually.
Real-world example: A backup cron. Runs on schedule, exits 0 or 1, logs to a file.

Arcanada context

See the registry list below for projects currently rated L1.

Projects currently rated at this level in the registry:

Partial

Closed-loop ops in narrow domain with static rules. Health endpoint, classified errors, post-run status to ops dashboard.

Human role: Supervises and approves; triages alerts.
Real-world example: Kubernetes liveness probe + auto-restart on failure within a fixed retry budget.

Arcanada context

See the registry list below for projects currently rated L2.

Projects currently rated at this level in the registry:

Verdicus now L2 → target L4 Auth Arcana now L2 → target L4 Transcribator now L2 → target L4 Arcanada Ops Bot now L2 → target L4 Muneral / Munera Arcana now L2 → target L3 Arcanada Support Center now L2 → target L3 Email Agent now L2 → target L3 Model Connector now L2 → target L3 Cubrim now L2 → target L5 Arcanada Ecosystem now L2 → target L4 Control Arcana now L2 → target L4

Conditional

Real-time sensing via structured observability (traces, eval scores). Heartbeat, post-deploy smoke gate, validated credentials, explicit exception hierarchy. Adapts via dynamic policies.

Human role: Validates non-trivial decisions; handles unknown failures.
Real-world example: AWS Auto Scaling driven by CloudWatch — a metric crosses threshold and the system scales out. The decision is policy-driven, the human approves the policy, not each action.

Arcanada context

Target level for Munera, Support, Scrutator, Model Connector, Agent Dreamer, Email Agent, ARCA Assistant.

Highly Autonomous · Self-Healing

Cross-domain self-detection, self-diagnosis, self-recovery. Retry-with-classification, circuit breakers, fallback chains, state checkpointing, recovery audit trail, known-fix mapping, post-recovery verification, hard cost circuit breakers for LLM calls.

Human role: Exception handling only — unknown-unknown failures.
Real-world example: Kubernetes self-healing (pod crash → reschedule), Netflix Chaos Monkey, Spinnaker auto-rollback. Plus, for AI agents: hard cost circuit breakers on LLM calls — a runaway loop without one can burn budget before a human notices.

Arcanada context

Target for Verdicus, Transcribator, Ops Bot (first self-healing pilot), Auth Arcana, Disk Arcana, Agent Dreamer.

Fully Autonomous · Self-Optimizing

Closed-loop across services. Proactive self-improvement: A/B testing prompts and models, dynamic model routing, cost governance with auto-downgrade, eval-driven prompt refinement, knowledge graph evolution.

Human role: Sets intent. No routine interaction.
Real-world example: A self-driving system tuning its own routing policies based on observed traffic patterns and a labeled outcome dataset.

Arcanada context

Strategic horizon — not a 2026 deliverable. Candidates: Transcribator (when volume crosses 1000+/day), Model Connector (multi-provider routing). Hard cost CB and labeled eval dataset are non-negotiable prerequisites.

Orchestrator Autonomy · Self-Assembly

The autonomous unit is no longer only an executor. An orchestrator can notice a recurring class of work, design a specialized agent, assign tools and memory, constrain permissions, define tests, run it in a sandbox, evaluate quality, plug it into the workflow, and roll it back if it fails.

Human role: Defines the boundary: which agents may be created, which tools and budgets are allowed, which tests are mandatory, and which actions still require approval.
Real-world example: A Datarim-like orchestrator that creates a new reviewer or publisher agent for a repeated workflow, verifies it against historical cases, and disables it when quality drops.

Arcanada context

Research horizon for Datarim, Consilium, Publisher, Control Arcana, and Managed Project Spaces. L6 is not "many agents"; it is autonomous design of an agent team under explicit limits.

Ecosystem Autonomy · Self-Development

The whole ecosystem can observe agent health, find bottlenecks, convene councils, compare architecture options, build eval datasets, maintain internal documentation, update its own rules, plan development, reallocate resources, and build tools for itself under human-set constraints.

Human role: Acts as legislator, not operator: mission, forbidden actions, budgets, legal and ethical constraints, escalation criteria, stop conditions, and review cadence.
Real-world example: An agent ecosystem that proposes a new service, runs a multi-role council, records objections and rollback plans, allocates a budget, implements in a sandbox, and asks for approval only at hard gates.

Arcanada context

Long-term Arcanada direction: an operational ecosystem with memory, councils, infrastructure loops, research, governance, and explicit rollback. This is not AGI; it is an autonomous engineering organization made of agents.

Why the upper pair exists

L6 and L7 measure not the executor, but the system that creates executors

L5 answers: how can the agent do its job better? L6 asks a different question: what new working unit should exist so this class of tasks is no longer handled manually? L7 raises the scale again: how does the whole ecosystem make decisions, build infrastructure, remember failures, and change its own rules without losing human-set boundaries? These levels do not call the system AGI; they make it engineerable.

Article: after self-healing

AI-specific dimensions

Operational dimensions without which L2–L7 become a declaration: cost, schemas, fallback, tool rights, and the eval loop.

LLM cost governance

A runaway LLM loop can burn budget before a human notices. Hard cost circuit breakers are non-negotiable.

Output schema validation

Prompt injection (OWASP LLM01) and schema drift in LLM output. Zod/JSON schema at L3, corrective retry at L4.

Model fallback chain

Provider outage / deprecation / rate limit demands runtime model swap (Claude → GPT → local). Single-model agents fail the L4 gate.

Tool-use scoping

Agentic file-write and shell-exec are a new attack surface (OWASP LLM06 Excessive Agency). High-risk tools are sandboxed at L4.

Eval / RLHF loop

Self-optimization without labeled eval has no signal. Every decision gets a downstream signal; the meta-loop tunes prompts.