AI Agent Operations

Autonomy is a tier, not a switch,
agents climb the ladder, one rung at a time

Agent Autonomy is the page where you pick what each agent is allowed to do without a human. Four tiers: shadow (logs only), suggest (writes runbooks), auto-rollback (acts but auto-reverts on regression), full auto. Promotions are slow, demotions are fast. Operators can override either way.

Get Started Talk to Sales
app.novaaiops.com / agent-autonomy
● LIVE
4
Autonomy tiers
Slow
promotion, fast demotion
Per-agent
config
Override
always available
The Four Tiers

Each tier is a precise contract

Shadow: agent observes, reasons, writes to memory; never proposes an action to a tool. Suggest: agent writes a runbook for a human to run; never executes itself. Auto-rollback: agent executes, but if Ground-Truth Verifier fails at T+5m, the platform automatically reverts. Full auto: agent executes without auto-revert, only used for low-risk actions on agents with high trust.

  • ✓
    Shadow: observe and learn, perfect for new agents in a new tenant
  • ✓
    Suggest: writes runbooks; the human is still the trigger; great for medium-risk classes
  • ✓
    Auto-rollback: acts; reverts if verifier fails at T+5m; the workhorse tier
  • ✓
    Full auto: acts and stays acted; used only for low-risk classes on agents with proven track record
app.novaaiops.com / agent-autonomy · tiers
Promotion Rules

Slow on the way up

Promotion requires three things: a minimum trust score (50 for suggest, 70 for auto-rollback, 90 for full auto), a minimum sample size (verified actions in the last 30 days), and an operator approval. There is no auto-promotion ever. Operators see a "ready for promotion" badge but the click is theirs.

  • ✓
    Trust floor per tier: 50/70/90 for suggest/auto-rollback/full auto, published, not negotiable
  • ✓
    Sample size minimum: at least 30 verified actions in the last 30d before promotion is offered
  • ✓
    Operator approval: no auto-promotion, every tier change is a human decision
app.novaaiops.com / agent-autonomy · promotion
Demotion Rules

Fast on the way down

Demotion is automatic and immediate. Three regressions in 7 days demote the agent one tier. A high-severity regression (caused or worsened a sev-1) demotes by two tiers. Trust drops below the tier floor demotes immediately. The asymmetry is intentional: trust takes weeks to earn, seconds to lose.

  • ✓
    Three regressions in 7d: → demote one tier · log to ledger · notify operator
  • ✓
    High-severity regression: → demote two tiers · page operator · ledger entry with explicit "high-sev demote" tag
  • ✓
    Trust below floor: → immediate demote regardless of timing
app.novaaiops.com / agent-autonomy · demotion
Per-Class Override

An agent can be at suggest globally but full-auto on one class

Sometimes an agent is great at one class and unproven at others. Per-class overrides let you set different tiers per class on the same agent. log-triager full-auto on "route alerts to teams" but only suggest on "summarize for postmortem." The override is captured in config and visible on the agent's autonomy detail page.

  • ✓
    Per-class tier: tier can be lower or higher than the agent's default per class
  • ✓
    Visible everywhere: the per-class tier shows on Agent Fleet, Agent Ledger, and Agent Audit
  • ✓
    Auditable: every override has an author, reason, and date, captured in the agent's ledger
app.novaaiops.com / agent-autonomy · per-class
Video walkthrough coming soon

Subscribe to Nova AI Ops on YouTube for demos, tutorials, and feature deep-dives.

Trust grows in steps

Adopting AI for ops is a series of small "let it do that" decisions. Autonomy tiers are how you make them deliberate.

Get Started Request a Demo