Shadow mode: the two weeks that make or break a launch

Every Vigil agent spends two weeks working beside your team before it works alone. It drafts, it proposes — but it doesn’t send. Those two weeks decide whether a launch sticks.

What we measure

Accuracy against what a human did, confidence calibration, and every case where the agent should have handed off but didn’t.

When to go live

• Accuracy above your threshold for five straight days.

• Zero unsafe actions proposed.

• The team says the drafts are usable as-is.

Shadow mode is where agents earn trust — one draft at a time.