Autonomy & guardrails

The autopilot is designed to earn trust incrementally. You decide how much it does on its own, and every mode runs inside guardrails you control — on top of the platform protections (frequency caps, quiet hours, suppressions) that always apply.

Autonomy modes

• Review required (default) — the autopilot classifies, detects, and drafts, then places each campaign in the review queue for a human to approve before it goes live. Best while you’re building confidence.
• Auto-execute — the autopilot also activates campaigns automatically, provided the opportunity fits within your guardrails. Anything that exceeds a limit falls back to review instead of being sent.

Guardrails

Guardrails bound what the autopilot may do. Set them during activation or adjust them any time.

• Max campaigns per day — a ceiling on how many campaigns the autopilot may launch in a day.
• Max recipients per day — a ceiling on total contacts messaged by the autopilot per day.
• Max audience per campaign — the largest a single auto-executed campaign’s micro-segment may be before it must fall back to human review.
• Minimum audience to engage — the autopilot won’t generate a campaign for a segment smaller than this. It defaults to a sensible floor (20 contacts) so effort isn’t spent on a handful of people.
• Participating channels — the autopilot may only send through the channels you selected, and only ones you’ve actually connected.

Cost controls

The autopilot is deliberately economical with the expensive part — the LLM used for campaign generation:

• Classification is deterministic and never uses the LLM.
• Each cycle caps the number of campaign generations; extra opportunities are recorded as detected and generate on a later cycle rather than all at once.
• Tiny segments are skipped via the minimum-audience threshold.
• Opportunities are de-duplicated per kind per day, so the same idea isn’t generated repeatedly.

When a cap or threshold drops work, it’s logged so the behavior is visible rather than silent.

Recommendation

Start in review-required mode. Once you’ve seen a few cycles of drafts and are comfortable with the quality and cadence, switch to auto-execute with conservative caps and widen them as results come in on the performance dashboard.

Next steps

• Adjust these settings in activation.
• Watch outcomes on the performance dashboard.
• Connect more channels to give the autopilot more reach.