Most organizations do not build all their AI agents in one place. Some run on Amazon Bedrock, some on Google Cloud, and many are built by developers with open-source frameworks such as LangChain or LangGraph. Microsoft Agent 365 can now discover agents on several of those platforms. But discovering an agent is not the same as governing it. Synergy Advisors experts have validated and extended Agent 365 so that agents outside Microsoft meet the same five conditions as Copilot: discovery, identity, security protection, data protection, and telemetry.
This article explains what Agent 365 covers out of the box for third-party agents, where the gaps are, and how we close them, platform by platform.
Why agents outside Microsoft matter now
Agent 365 has been generally available since November 3, 2026, and it is Microsoft’s centralized portal where IT, security, and compliance teams manage AI agents. The problem is that the agents themselves are spread across clouds and frameworks. A data science team builds on Amazon Bedrock, a product team uses Google’s Agent Development Kit (ADK), and developers ship LangChain or LangGraph agents from their own pipelines. If only the Microsoft-built agents are governed, the rest become the shadow AI agents that nobody owns.
The principle we apply is simple: innovation stays distributed; trust, policy, and accountability stay centralized. Teams keep building where they work best. Governance happens in one place.
What Agent 365 covers out of the box
Agent 365 brings external agents in through connected platforms, also called registry sync. At the time of this writing, Microsoft supports seven: Amazon Bedrock, Google Vertex AI, Salesforce Agentforce, Databricks Genie, Anthropic Claude (Managed Agents), Oracle Generative AI Agents, and Snowflake Cortex. Registry sync adds those agents to the Microsoft 365 agent registry and, where the platform supports it, shows session, exception, runtime, and tool-call metrics on each agent’s Activity tab. Observability data collection is part of the Frontier preview program.
Registry sync has limits: it does not add code-level capabilities. Identity, observability, tooling, and notifications come from the Agent 365 SDK, which has to be integrated into each agent. Agent 365 also never invokes the synced agents or writes to the source platform. In practice, an agent on Bedrock can appear in your inventory and still have no Entra identity, no data loss prevention on its prompts, and no usage data your teams can act on.
The five conditions, platform by platform
We use five conditions to evaluate third-party agents. The table below shows what Agent 365 provides out of the box today, and what Synergy Advisors has validated and extended:
| Platform | Discovery | Identity | Security protection | Data protection | Telemetry |
|---|---|---|---|---|---|
| Amazon Bedrock AgentCore (AWS) | Out of the box | Not covered | Partial | Not covered | Partial |
| Google ADK (Gemini Enterprise Agent Platform) | Out of the box | Not covered | Partial | Not covered | Partial |
| LangChain | Not covered | Not covered | Not covered | Not covered | Not covered |
| LangGraph | Not covered | Not covered | Not covered | Not covered | Not covered |
| With Synergy Advisors | Validated and extended (Agent 365 SDK) | Validated and extended (Agent 365 SDK) | Validated and extended (Agent 365 SDK) | Validated and extended (Purview APIs) | Validated and extended (Agent 365 SDK) |
- Out of the box: native in Agent 365 through connected platforms (registry sync).
- Partial: read-only signals only, such as telemetry and Defender alerts where the platform reports them. No code-level controls.
- Not covered: needs code-level integration with the Agent 365 SDK.
- Validated and extended: integrated and tested by Synergy Advisors with the Agent 365 SDK, Microsoft Entra Agent ID, Microsoft Purview, and OpenTelemetry.
The result: every condition, on every platform in the table, is covered.
What we extend, condition by condition
| Condition | Out of the box | What our experts add | How we build it |
|---|---|---|---|
| Discovery | AWS and Google Cloud agents synced into the agent registry | Every agent, on any platform, onboarded with an accountable owner, a sponsor, and an approved blueprint | Agent 365 SDK, agent blueprints |
| Identity | Read-only registry entry; the platform connection uses a key or service account | Each agent gets its own Microsoft Entra Agent ID, a sponsor, and risk-based Conditional Access. See AI Agent Identity | Entra Agent ID, Agent 365 SDK |
| Security protection | Agent telemetry feeds Defender alerts and insights | Agent operations in Defender Advanced Hunting, correlated in Microsoft Sentinel for one SOC response | Advanced Hunting, Sentinel analytics |
| Data protection | Not included: registry sync adds no code-level controls | Prompts and responses evaluated by Microsoft Purview DLP in real time; sensitive content is blocked before it leaves | Microsoft Graph processContent, Entra app identity |
| Telemetry | Session, runtime, and tool-call metrics on the Activity tab | Who uses each agent, how often, and how much, sent back to Agent 365 dashboards and logs | OpenTelemetry, Agent 365 SDK |
Telemetry deserves a note: secured is not the same as measured. An agent can be protected by Entra, Purview, and Defender and still be invisible to the business. We instrument each agent with OpenTelemetry, map its spans and metrics to Agent 365 observability, and prove it live in the Activity tab, dashboards, and logs.
On AWS: from visibility to enforcement
For Amazon Bedrock AgentCore, we work in two levels. Your agents stay on Bedrock: nobody has to move them or choose one cloud over another.
| Level 1 · Connected visibility | Level 2 · Deep integration | |
|---|---|---|
| What you get | The agent appears in Agent 365 (platform: Amazon Bedrock). Sessions and activity are synced from Amazon CloudWatch. Agent operations are available in Defender Advanced Hunting. | Prompts and responses are evaluated by Purview in real time. Sensitive content is blocked and safe prompts are answered normally. DLP events appear in Activity explorer, correlated with CloudWatch and Defender. |
| What it needs | AgentCore runtime ready with observability enabled, CloudWatch Transaction Search turned on, and a connected platform with least-privilege, read-only IAM access. | An Entra app registration with the Microsoft Graph processContent permission, a Purview DLP policy scoped to that app identity, and the agent runtime configured with the Purview integration. |
On Google Cloud and open-source frameworks
- Google Cloud. Agents on Gemini Enterprise Agent Platform are imported by project and region into the agent registry. With ADK callbacks, prompts and responses go to Purview before the model answers, and Cloud Trace spans, BigQuery trace exports, and Cloud Monitoring metrics are mapped to Agent 365 through OpenTelemetry. Google agents stay on Google Cloud and join the same control plane for identity, data, security, and usage.
- LangChain and LangGraph. These frameworks have no connected platform, so nothing is covered until the agent is integrated. We register each agent with the Agent 365 SDK, give it an Entra identity and an owner, and bring it under the same observability, security, and lifecycle controls as a Copilot agent.
For IT, the outcome is one inventory of every agent. For security, one SOC view, on any cloud. For compliance, one audit trail. For the business, real data on adoption and usage.
What comes next
The four platforms above are available today. Our experts are defining Agent 365 SDK integrations for more agent platforms, and a managed API gateway in Azure to govern the MCP servers your agents use, with policy and visibility in one place. The principle does not change: innovation stays distributed, control stays centralized.
How our experts help
- Agent 365 E-xperiences. From an Architecture Design Session that maps every agent you run to the five conditions, to a proof of concept and a production pilot on your own platforms.
- E-mmersion Dev Track. A hands-on, instructor-led lab of up to four hours for developers and platform engineering teams (level 400), in English or Spanish. Your team onboards LangChain, LangGraph, and Google ADK agents with the Agent 365 SDK, applies Entra Agent ID and least-privilege access, and reviews observability and lifecycle, in a preconfigured lab with no setup in your own tenant.
On licensing: onboarding third-party agents uses the Agent 365 SDK, and it assumes the Agent 365 prerequisite is already met, Microsoft 365 E5 with the Agent 365 add-on, or Microsoft 365 E7.
Frequently asked questions
Can Agent 365 govern agents that run on AWS or Google Cloud?
It can discover them through connected platforms and show their activity where the platform supports it. Identity, data loss prevention, and full telemetry need code-level integration with the Agent 365 SDK, which our experts validate and extend.
Do we have to move our agents to Azure?
No. Agents stay on Amazon Bedrock, Google Cloud, or wherever your developers build them. Only the governance moves to Agent 365.
What does registry sync not cover?
According to Microsoft Learn, registry sync does not add code-level capabilities such as identity, observability, tooling, or notifications, and Agent 365 never invokes the synced agents.
Which platforms and frameworks do your experts cover?
Amazon Bedrock AgentCore, Google ADK, LangChain, and LangGraph today. Integrations with more platforms are in the works!
Next step
Synergy Advisors experts can map the agents you already run, on any cloud, to the five conditions and show you what it takes to govern them in Agent 365. Join the Beyond the Agent webinar on Tuesday, November 17, at 2:00 p.m. ET, or talk to our experts.
Sources
- Registry sync in the Microsoft 365 agent registry Microsoft Learn, updated November 6, 2026
- Connect existing agents to Agent 365 Microsoft Learn
- Microsoft named a Leader in the 2026 Gartner Magic Quadrant for AI Agent Management Platforms Microsoft Security Blog, November 3, 2026
- Agent 365 (A365) E-xperiences Portfolio, V.1.7 Synergy Advisors, November 2026
