What does the AI do?
Start with a task and a measurable outcome. We distinguish generated content, retrieved information, and automated actions on each product page.
Our AI technology approach
Useful AI starts with a clear product workflow.
Our software brings together a focused interface, a practical task, and the AI capabilities that fit it. Microsoft Foundry on Azure is the intended platform for most of our future AI-enabled releases.
What information can it access?
Explain supported inputs, connected sources, access requirements, and data boundaries. A model's general knowledge is different from your organisation's records.
What actions can it take?
Identify tools and integrations, the permissions they need, and the points where a person reviews or approves an action.
How will quality be assessed?
For future AI releases, we intend to use representative tasks, stated evaluation criteria, and documented limitations. Product pages will distinguish planned tests from published evidence.
The platform behind our development direction
Microsoft Foundry building blocks.
Microsoft Foundry, formerly Azure AI Foundry, brings together models, agents, tools, knowledge retrieval, and AI evaluation on Azure. It is our intended AI platform for most future products. Each product page identifies the services actually used and their implementation status.
Platform information reviewed . Feature, region, model, and API availability can change; consult the linked Microsoft documentation.
Foundry Models
Choose AI suited to the task
Model access and supported customisation provide options for text, reasoning, and multimodal applications. The product design determines which model fits its quality, latency, and cost needs.
Potential product fit: summarising documents or assisting a decision with generated text.
Core model flows are generally available; individual models and deployment types have their own availability and terms.
Read Microsoft's documentationFoundry Agent Service
Connect AI to a useful workflow
Prompt agents and hosted custom agent code can combine instructions, models, and tools. A product still needs clearly defined permissions, action boundaries, and human review points.
Potential product fit: an assistant that retrieves information and proposes a next action.
Core agents are generally available. Voice-based experiences and several advanced capabilities remain in preview.
Read Microsoft's documentationFoundry IQ
Answer from relevant business information
Knowledge bases and agentic retrieval can gather context from supported sources and return citations. Permission enforcement depends on the chosen source, identity setup, and integration.
Potential product fit: finding an answer across a connected document collection with source references.
Mixed availability: some retrieval APIs are generally available; portal access and other features remain in preview.
Read Microsoft's documentationFoundry Tools and toolboxes
Work with the right inputs and systems
Tools, custom APIs, and MCP connections can extend agent workflows. Speech, translation, and document-processing services provide further building blocks when a product needs them.
Potential product fit: analysing a document, translating content, or requesting an action through an approved integration.
Tools and toolboxes have generally available core experiences; check the status of each tool and connected service.
Read Microsoft's documentationEvaluations and observability
Measure quality and understand behaviour
Evaluation datasets, tracing, and red teaming can reveal failures and compare releases. Useful evidence comes from representative product tasks and clearly stated evaluation criteria.
Potential product fit: testing whether answers are relevant, grounded, and suitable for the intended workflow.
Core evaluations and selected tracing are generally available. Monitoring and some evaluators and analysis features remain in preview.
Read Microsoft's documentationFoundry Control Plane and identity
Define access, oversight, and operating boundaries
Microsoft Entra identity, roles, policies, and management tools can support oversight of agents, models, and tools. Their presence in Azure does not establish that a product has configured them or achieved a compliance certification.
Potential product fit: administering an AI application with explicit roles and operational visibility.
Core identity and project controls are available; several Control Plane operating and governance experiences remain in preview.
Read Microsoft's documentationWhat is available in our products?
Platform possibilities and shipped product capabilities are different. Check the implementation status and limitations for the product you want to use.
Interactive prototype
AI Delivery Dashboard
This prototype helps teams explore the management of AI work. It does not currently generate AI responses, run agents, or connect to Microsoft Foundry.
Read product capabilities