How to Translate AI Tunables into Business Workflow Permissions

In Verdify Lab, Iris does not receive an open-ended instruction to run the greenhouse. It works through tunables: bounded fields that describe tactical intent and can be validated before they affect the physical system.

That pattern translates directly to business workflows.

From tunables to permissions

In a support workflow, the tunables might be ticket tags, urgency, routing recommendations, draft replies, or escalation notes.

In a quality workflow, they might be evidence packet summaries, classification suggestions, missing-document flags, or reviewer checklists.

In a CPG workflow, they might be retailer response drafts, chargeback categories, claim evidence summaries, or exception queues.

The important move is to define the permission type:

• Read: the AI can inspect approved sources.
• Classify: the AI can apply labels or routing suggestions.
• Draft: the AI can prepare text for review.
• Recommend: the AI can propose a next step.
• Execute: the AI can act only when explicitly allowed, reversible, and logged.
• Prohibited: the AI must not touch the action.

Translate this to your workflow

Do not start with “Should we use an agent?” Start with “What fields may the agent write, and what layer decides whether those writes count?”

That is the work of the Boundary Matrix, the AI Workflow Boundary Matrix, and the first phase of a Bounded AI MVP Sprint.