GOVERNANCE
Autonomous where it makes sense. Human where it matters.
AI does the work. Humans approve the consequences.
PRINCIPLES
How we think about agent governance
Seven principles guide every agent design.
Human oversight
Every agent operates within defined boundaries. Consequential actions are routed to the appropriate person for review and approval before they execute.
Least-necessary access
Agents are configured with the minimum permissions required to perform their work. Access is scoped to specific systems, data sets and action types.
Approval points
The design phase defines where human approval is required based on the action type, risk level and your business rules. These controls are configured before the agent goes live.
Auditability
Agent actions can be reviewed and traced where supported by the underlying platform. Governance reviews are part of ongoing AgentOps.
Operating boundaries
Each agent has clear instructions defining what it can and cannot do. These boundaries are refined over time as you build confidence in the agent.
Testing protocols
Agents are tested against real scenarios before going live. Governance controls are validated as part of the deployment process.
Your environment, your control
Agents run inside your own Microsoft 365 tenant. Governance controls, data and audit trails stay within your environment - not ours.
WHY IT MATTERS
What happens without governance
These are the risks clients already worry about. Good governance addresses each one.
Agent drift
Without regular reviews, agents gradually deviate from intended behaviour as business context changes around them.
Permission creep
Over time, agents can accumulate more access than they need, creating unnecessary exposure.
Blind spots
Without monitoring, there is no visibility into what agents are actually doing or how they are performing.
Missing audit trails
If agent decisions are not logged, there is no way to review, explain or defend what happened.
Knowledge decay
Agents referencing outdated information make confident but incorrect decisions.
Uncontrolled actions
Without clear boundaries, agents may take actions beyond what was intended or appropriate.
PROGRESSIVE MODEL
Progressive governance
Not all agent actions carry the same weight. Governance scales with consequence.
Routine actions
Read, search, classify, summarise, compare, prepare, reconcile, flag
These can often happen automatically.
Controlled actions
Create internal tasks, update certain records, request missing information, prepare documents, route work
Controls depend on your business policy.
Consequential actions
Important external communications, financial approvals, payments, contractual actions, sensitive record changes, regulatory submissions
Routed to the appropriate person for approval.
This is Ditto's design philosophy, not legal advice or a universal compliance requirement.
IN PRACTICE
Governance in practice
What actual governance activities look like across your agents.
Permission audits
Review what each agent can access and ensure permissions are still appropriate for its current role.
Boundary reviews
Check that operating limits match how the business currently works and adjust where needed.
Action log reviews
Examine agent decisions and outcomes to identify patterns, errors or unexpected behaviour.
Escalation testing
Verify that approval workflows trigger correctly and reach the right people.
Knowledge checks
Confirm agents are referencing current, accurate information sources.
Change assessments
Evaluate how business changes - new staff, updated processes, system updates - affect agent behaviour.
QUESTIONS
Common questions
Who is responsible for agent governance?
Ultimately, your business. Ditto designs the governance controls, manages the reviews and handles the day-to-day oversight - but accountability for agent behaviour stays with you. We make that easy to maintain.
How often are governance reviews conducted?
It depends on the agent and its risk profile. Higher-consequence agents are reviewed more frequently. Reviews are part of AgentOps, not a separate engagement.
What happens when a governance issue is found?
We flag it, explain the risk, and recommend a fix. If it is urgent, we act immediately. If it requires a decision from you, we provide the context you need to decide quickly.
Do we need a governance framework before deploying agents?
No. Ditto builds governance into the agent design process. You do not need a separate framework - governance controls are part of every agent we deploy.
Where do our agents and data live?
Inside your own Microsoft 365 tenant. Ditto manages your agents, but the environment, data and audit trails are yours. Nothing leaves your control.
Can governance controls be adjusted over time?
Yes. Governance should evolve as you build confidence in your agents. Controls that start strict can be relaxed as the agent proves reliable, and new controls can be added as needs change.
Is governance included in AgentOps?
Yes. Ongoing governance reviews are a core part of AgentOps. If you are on an AgentOps plan, governance is covered.
Ready to discuss governance for your agents?
Talk to Ditto about how we design governance controls that fit your business.