A woman monitors a glowing AI workflow dashboard with interconnected bots, data systems, and security controls.
Hewlett Packard Enterprise is extending ChatHPE from an internal knowledge assistant into a platform for governed AI actions. Microsoft’s customer account describes HPE-developed AgentOps technology supporting hundreds of production agents—a shift from finding information to managing software that can act on it.

For enterprise IT teams, the distinction matters: a useful answer and an authorized action are different engineering problems.

A woman monitors a glowing AI workflow dashboard with interconnected bots, data systems, and security controls. What HPE has built​

According to Microsoft, approximately 30,000 employees use ChatHPE, and more than 300 applications consume its capabilities. Uses include HR questions, technical research, hardware information and knowledge synthesis. Reported time savings are not quantified.

HPE Distinguished Technologist Jose Mejias describes an October-to-February development cycle, with a prototype ready in under three months; the account does not identify the years. ChatHPE uses Foundry IQ to ground answers in authorized enterprise content and provide citations.

Those figures establish reported reach, not independently measured productivity. Nor should application consumption be confused with a count of autonomous agents.

AgentOps adds an operating layer​

HPE’s AgentOps handles agent registration, discovery, orchestration, monitoring and management. Its foundation includes Microsoft Entra ID for identity, Azure Kubernetes Service for orchestration and Defender for Cloud for security. Copilot Studio agents can participate alongside other technologies.

HPE also describes interoperability involving Microsoft Foundry, Copilot Studio and Agent2Agent, or A2A. Existing systems remain authoritative. Workload placement considers security, sensitivity, performance, cost, capacity and governance; Azure is often the starting point, not necessarily every workload’s destination.

The architectural lesson is separation of responsibilities. A conversational interface can collect intent, retrieval can supply supporting information, and an operational layer can coordinate execution. Treating those as distinct responsibilities makes a more useful design discussion than simply asking which model powers the chatbot.

Permission-aware retrieval is not automatic​

Microsoft Learn supplies an important qualification beyond the customer account: Foundry IQ requires Azure AI Search, but does not require Foundry Agent Service. Applications can call its knowledge-base APIs directly.

More importantly, Microsoft’s Foundry IQ FAQ says user-level access controls depend on the knowledge source supporting document-level permissions and on those permissions being explicitly configured for synchronization. Administrators should not assume that connecting a repository automatically preserves every document’s access restrictions.

That is a product-level boundary, not evidence of a flaw in HPE’s implementation. The practical takeaway is to validate authorization source by source rather than treating “permission-aware” as a universal switch.

What administrators should take away​

As an architectural assessment—not a description of undisclosed HPE settings—teams evaluating a similar transition should ask:

  • Retrieval: Can users obtain restricted information through answers, snippets or citations?
  • Execution: Which identity authorizes an action, and what limits its authority?
  • Oversight: Which actions require approval, and how can execution be stopped?
  • Evidence: Can operators reconstruct the request, retrieved context, tool invocation and outcome?
  • Recovery: What happens when an action fails halfway through?

HPE’s account offers a concrete enterprise example, but not a reproducible deployment blueprint: it provides neither named agent workflows nor detailed approval flows or quantified operational outcomes.

The useful conclusion is narrower than “agents are ready for everything.” Moving beyond answers calls for a separate examination of authority, observability and recovery. A citation can help justify an answer; it is not, by itself, permission to change a business system.

 

References

  1. HPE moves from trusted AI answers to governed action with Azure - Microsoft Microsoft Tue, 17 Nov 2026 00:15:17 GMT
  2. Foundry IQ FAQ - Microsoft Foundry | Microsoft Learn learn.microsoft.com
  3. What is Foundry IQ? - Microsoft Foundry | Microsoft Learn learn.microsoft.com