Human-in-the-Loop Design Pattern
High-stakes, ambiguous, or irreversible actions are reviewed by a human before execution, via Skill Call Approvals or Business Approvals.
Why This Pattern
The split between Skill Call Approvals and Business Approvals tracks what the approval is actually protecting against. Skill Call Approvals exists because a permission gap exists — the acting user can't do this themselves, so someone with privilege has to sign off. Business Approvals exists because the action is consequential enough that a human should confirm it, independent of anyone's permissions. Conflating the two either annoys users with approvals that aren't really about permission, or skips a genuinely necessary permission check.
When to Use It
- User-scoped connections aren't possible for the action, but the underlying API supports acting on a user's behalf.
- The action is irreversible, high-value, or compliance-sensitive.
When NOT to Use It
- Routine, low-risk actions with high confidence — gating every action trains reviewers to rubber-stamp.
- The workflow runs headless (via Assign Task to Genie) — Business Approvals can't be used there, since approvals are delivered through a chat interface a headless call doesn't have.
How It Works in Workato
Skill Call Approvals: put skill execution behind approval so the action only runs once a privileged user approves it. Business Approvals: the mechanism for a recipe-orchestrated workflow to receive human sign-off and notify users directly through their chat interface.
An approval request needs to include what's being proposed, why, the specific data involved, and the consequences of approval vs. rejection — a request without enough information results in rubber-stamping.
Source: Agentic Design Cookbook V2.0, p. 45 (Skill Call Approvals), p. 28 (Approval for Skill Call) · Workato Docs: Business Approvals