This guide helps you configure AI at the workspace level before teams start using AI broadly across authoring, automation, and reader experiences. It is intended...
This guide helps you configure AI at the workspace level before teams start using AI broadly across authoring, automation, and reader experiences. It is intended...
This guide helps you configure AI at the workspace level before teams start using AI broadly across authoring, automation, and reader experiences. It is intended for admins and leads who want AI to be useful, safe, and cost-controlled from the start.
Who this is for
- workspace owners and admins
- teams rolling out AI to several projects
- organizations that need policy, budget, and terminology consistency
What the workspace AI policy controls
The AI Policy surface is divided into:
- Models & Tiers
- Token Budgets
- Feature Controls
- Agent Autonomy
- Style & Terminology
- Data Boundaries
Treat this as a governance surface, not a tuning playground. The right goal is a policy that supports real work while making unsafe or inconsistent behavior less likely.
Step 1 — Choose the allowed models and default tier
Start with Models & Tiers.
Set:
- which models are enabled
- the default response tier
- feature-specific overrides where needed
Use a smaller model set unless you have a clear operational reason to expose every option. More choices increase flexibility, but they also increase inconsistency and policy surface area.
Step 2 — Set token budgets before usage scales
In Token Budgets, configure:
- the monthly workspace budget
- any per-member or project-level caps you want to enforce downstream
- alert thresholds and alert recipients
This step is easiest before teams form habits around unrestricted usage.
Step 3 — Decide which AI features are allowed
In Feature Controls, review each enabled surface independently.
Common examples include:
- workspace assistant
- writing or authoring assistance
- improve-with-AI style workflows
- scan or quality analysis
- site assistant
- agent capability
Do not assume all teams need every feature turned on immediately. It is often better to enable the lowest-risk surfaces first, then expand once you trust the operating model.
Step 4 — Set the autonomy ceiling
In Agent Autonomy, decide how much independent action AI is allowed to take.
Typical operating modes:
- suggest only
- draft for review
- autonomous
For most documentation teams, start with a review-centered posture. Let AI produce drafts and suggestions before it is allowed to act without human review.
Step 5 — Codify style and terminology
The Style & Terminology section is one of the most undervalued parts of AI policy.
Use it to define:
- voice
- person
- reading level
- preferred terms
- banned terms
This reduces the drift that happens when different AI surfaces generate content in slightly different tones or with different naming conventions.
Step 6 — Define data boundaries explicitly
Use Data Boundaries to set the privacy posture that applies to AI usage in the workspace.
Review options such as:
- no-training posture
- zero-retention options
- data residency preferences
- redaction or PII handling patterns
The best settings here depend on your organization’s actual requirements, but the important part is that they are consciously chosen and documented.
Step 7 — Publish the policy operationally, not just technically
Once the policy is configured:
- tell project owners what is enabled
- explain which features are intentionally off
- explain the approval model
- explain budget expectations
- explain terminology rules that AI will follow
This avoids the common failure mode where the policy exists, but teams do not understand how it should shape their behavior.
Suggested rollout pattern
For most teams:
- enable a conservative model set
- set a real monthly budget
- enable authoring assistance and scans first
- keep automation and autonomy conservative
- codify terminology before generated content scales
- expand only after observing usage patterns