Microsoft Foundry: Building Enterprise AI Agents

Microsoft has consolidated its enterprise AI capabilities under a single unified platform: Microsoft Foundry. Formerly known as Azure AI Foundry, this rebrand reflects a broader ambition — to give organisations everything they need to build, deploy, and govern AI applications and autonomous agents at scale.

Whether you are looking to deploy a customer service agent, ground AI responses in your proprietary data, or connect models to 1,400+ business systems via the Model Context Protocol, Foundry provides the tools and governance framework to do it responsibly. In this guide, we explore what Microsoft Foundry offers, how its components fit together, and where it delivers practical value for enterprises.

What Is Microsoft Foundry?

Microsoft Foundry is an end-to-end AI development platform that brings together model selection, agent building, data grounding, and enterprise governance in a single environment. It replaces the fragmented landscape of Azure AI Studio, Azure Machine Learning, and various Cognitive Services with one cohesive experience.

At its core, Foundry is designed around three principles:

Choice — access to thousands of models from multiple providers, not just Microsoft's own.
Integration — deep connections to the Microsoft ecosystem, including Dynamics 365, Power Platform, and Microsoft Fabric.
Trust — built-in responsible AI controls, content safety filters, and enterprise-grade security.

For organisations already invested in the Microsoft stack, Foundry represents a natural extension of their existing infrastructure. For those evaluating AI platforms more broadly, it offers one of the most comprehensive model catalogues available today. You can learn more about our Microsoft Foundry consulting services.

The Model Catalogue: 11,000+ Models

One of Foundry's most compelling features is its model catalogue, which now includes over 11,000 models from a wide range of providers. This is not a Microsoft-only affair. The catalogue spans:

OpenAI — GPT-4o, GPT-4.1, o3, and the full range of reasoning models.
Meta — Llama 4 Scout, Llama 4 Maverick, and earlier Llama variants.
Mistral — Mistral Large, Medium, and specialised coding models.
DeepSeek — DeepSeek-R1 and other open-weight reasoning models.
Google — Gemma models available for deployment.
Cohere, Anthropic, and others — a growing roster of third-party providers.

The catalogue supports multiple deployment options. You can use Models as a Service (MaaS), where you pay per token with no infrastructure management, or deploy models on dedicated compute for greater control and data isolation. Foundry also supports fine-tuning, distillation, and custom model evaluation, so you can tailor models to your specific domain.

Benchmarking and Evaluation

Foundry includes built-in evaluation tools that let you compare models against your own test datasets. Rather than relying on generic benchmarks, you can measure accuracy, latency, and cost against the queries that matter to your business. This is particularly valuable when deciding between, say, a large reasoning model and a smaller, faster alternative for a specific task.

Foundry Agent Service: Building Autonomous AI Agents

The Foundry Agent Service is where Microsoft's vision becomes most ambitious. Rather than simply exposing models via an API, Foundry provides a framework for building autonomous agents — AI systems that can plan, reason, use tools, and take actions across multiple steps.

An agent built on Foundry can:

Plan and decompose tasks — breaking a complex request into a sequence of actions.
Call external tools and APIs — retrieving data, updating records, or triggering workflows.
Maintain memory and context — remembering prior interactions and building on them.
Collaborate with other agents — multi-agent architectures where specialised agents work together.

This goes well beyond simple chatbot functionality. A Foundry agent can, for example, receive a customer complaint, look up the customer's order history in Dynamics 365, check inventory levels, draft a response, and escalate to a human if the issue exceeds its authority — all autonomously.

Multi-Agent Orchestration

Foundry supports multi-agent patterns where a supervisory agent delegates tasks to specialised sub-agents. This is particularly useful for complex workflows where different agents handle different domains — one for data retrieval, another for analysis, and a third for customer communication. The orchestration layer manages handoffs, conflict resolution, and fallback behaviour.

Foundry IQ: Grounding AI in Enterprise Data

Foundry IQ addresses one of the most persistent challenges in enterprise AI: ensuring that model responses are grounded in your organisation's actual data rather than generic training knowledge.

Foundry IQ connects to your data wherever it lives — SharePoint, OneDrive, Microsoft Fabric, Azure SQL, Dataverse, and third-party sources. It creates a semantic index that agents and applications can query, ensuring responses reflect your current business reality.

This is not simply retrieval-augmented generation (RAG), though RAG is part of the picture. Foundry IQ also supports:

Knowledge graphs — structured representations of entities and relationships within your data.
Real-time data access — agents can query live data sources rather than relying on stale indexes.
Permission-aware retrieval — results respect your existing access controls, so users only see data they are authorised to access.

For regulated industries — finance, healthcare, legal — this permission-aware grounding is essential. It means you can deploy AI assistants that reference internal policies and customer records without risking data leakage across organisational boundaries.

MCP Support: 1,400+ Business System Connections

The Model Context Protocol (MCP) is an open standard that allows AI models and agents to connect to external tools and data sources through a standardised interface. Microsoft has embraced MCP within Foundry, providing pre-built connectors to over 1,400 business systems.

This means a Foundry agent can interact with:

CRM systems — Salesforce, HubSpot, Dynamics 365.
ERP platforms — SAP, Oracle, NetSuite.
Productivity tools — Jira, Confluence, ServiceNow, Slack.
Data platforms — Snowflake, Databricks, Google BigQuery.
Custom APIs — any system with a REST or GraphQL endpoint.

If you are already familiar with MCP from other contexts — such as our guide on using MCP for SEO tool integration — the concept is the same, but applied at enterprise scale. Foundry's MCP implementation handles authentication, rate limiting, error recovery, and audit logging, which are critical requirements for production deployments.

The practical benefit is significant: rather than building bespoke integrations for each system your agent needs to access, you configure MCP connections once and any agent in your Foundry environment can use them.

Enterprise Governance, Responsible AI, and Security

Microsoft has invested heavily in governance and responsible AI tooling within Foundry, and for good reason. Enterprises deploying autonomous agents need assurances around safety, compliance, and auditability.

Content Safety and Guardrails

Foundry includes Azure AI Content Safety, which provides real-time filtering for harmful content, prompt injection attacks, and jailbreak attempts. You can configure safety thresholds per application, so a customer-facing chatbot might have stricter filters than an internal research tool.

Audit and Compliance

Every agent interaction can be logged, traced, and audited. Foundry provides:

Prompt and response logging — full records of what the model received and returned.
Tool call tracing — which external systems were accessed and what data was retrieved.
Decision audit trails — for multi-step agent workflows, a complete record of the reasoning chain.

Data Residency and Encryption

Foundry supports regional data residency requirements, with deployment options across Azure's global regions. Data is encrypted at rest and in transit, and customer-managed encryption keys are supported for organisations with strict key management policies.

Integration with the Microsoft Ecosystem

Foundry does not exist in isolation. Its deepest integrations are with the broader Microsoft platform:

Dynamics 365 — AI agents can access customer records, sales pipelines, and service cases directly. For a deeper look at how AI is transforming Dynamics 365, see our Dynamics 365 AI and Copilot guide.
Power Platform — Copilot Studio, Power Automate, and Power Apps can all consume Foundry models and agents, enabling citizen developers to build AI-powered workflows without writing code.
Microsoft Fabric — Foundry IQ connects natively to Fabric lakehouses and warehouses, enabling agents to query analytical data at scale.
Microsoft 365 Copilot — custom agents built in Foundry can be surfaced as Copilot extensions within Teams, Outlook, and other Microsoft 365 applications.

This ecosystem integration is arguably Foundry's strongest differentiator. If your organisation already runs on Microsoft infrastructure, the path from pilot to production is considerably shorter than with a standalone AI platform.

Practical Use Cases

Customer Service Agents

Deploy agents that handle tier-one support queries autonomously. The agent accesses customer records via Dynamics 365, checks knowledge bases for solutions, and escalates to human agents when needed — all with full audit trails.

Data Analysis and Reporting

Agents connected to Microsoft Fabric can answer natural language questions about business data, generate reports, and flag anomalies. Finance teams can ask, "What were our top-performing products in Q4?" and receive an accurate, sourced answer.

Process Automation

Combine Foundry agents with Power Automate to create intelligent automation workflows. An agent can monitor incoming invoices, validate them against purchase orders, flag discrepancies, and route approvals — reducing manual processing time by up to 80%.

Knowledge Management

Ground agents in your SharePoint document libraries and internal wikis. Employees can ask questions about company policies, technical procedures, or project documentation and receive answers that cite specific source documents.

Getting Started

Microsoft Foundry is available through Azure, with pricing based on model usage, compute, and the specific services you deploy. Free tiers are available for experimentation, and enterprise agreements provide committed-use discounts.

For organisations looking to evaluate Foundry or build their first AI agents, Innotek Solutions provides consulting, architecture design, and implementation support. We can help you navigate the model catalogue, design agent workflows, and establish the governance frameworks needed for responsible deployment.

Ready to explore what Microsoft Foundry can do for your organisation? Get in touch with our team to discuss your requirements and start planning your enterprise AI strategy.