Are we governing agents, or just scripting them more carefully?

With multi-agent work, I kept seeing two extremes: script every step, or give agents a goal and watch the original intent get lost between handoffs.

Permissions and approval gates govern actions. The harder question is what an agent is allowed to decide while doing the work.

We now give agents a mandate: the outcome, scope, rules, ownership, escalation points and evidence required. They can reason freely inside those boundaries.

A workflow says what happens next. Governance says what the agent is allowed to decide.

How are you defining that boundary today?