Horizon for AI Agents
Validation in the Microsoft Agent 365 era.
The new validation surface inside Dynamics.
Microsoft Agent 365 went GA on May 1, 2026. Copilot Studio agents are landing inside D365 Sales, Service, Finance, Supply Chain, and Commerce. They take actions on enterprise data. They will pass your governance review under Agent 365. They will not pass the audit unless someone tests what they actually do.
Governance is not validation.
Agent 365 tells you which agents exist, what they have access to, and whether they are behaving safely at the identity and network layer. It does not tell you whether the agent's outputs are right. Validation answers the harder question: when this agent runs in production, does it do what the business needs?
How Horizon validates Copilot agents.
Horizon for AI Agents treats agents as workflow participants. Same deterministic test framework. Same audit log. Same suites that already cover your D365 environment.
- Agent inputs, verified against expected schemas and security boundaries
- Agent outputs, verified against business rules, not just structural validity
- Agent decisions, captured, replayable, comparable across releases
- Surrounding workflow, tested end-to-end via End-to-End Workflows
The risk profile of un-validated agents.
An agent inside D365 Finance can post journal entries. An agent inside Supply Chain can approve purchase orders. An agent inside Sales can change opportunity stages. Each agent decision becomes part of the audit trail your CFO, controller, or auditor will eventually have to defend. Validation is the protection.
