Back to Blog

Microsoft Foundry - A Practical Guide for Australian Businesses Getting Started

March 9, 2026•8 min read•Team 400

If you've been building AI projects on Azure over the past two years, you've probably noticed the platform shifting under your feet. Azure OpenAI Service, Azure AI Studio, Azure AI Foundry - the branding has moved fast. The latest iteration is Microsoft Foundry, and despite the name changes, this one feels like it's sticking around. More importantly, it's actually good.

We've been working with the platform since the early Azure AI Studio days, and the current version is the first time it feels like a genuinely unified environment rather than a collection of services stitched together with a portal. For Australian businesses running AI projects on Azure, understanding what Foundry actually does - and what it doesn't - is worth your time.

What Microsoft Foundry Actually Is

Microsoft Foundry is Azure's platform for building, deploying, and managing AI applications and agents. Microsoft calls it "the AI app and agent factory," which is marketing-speak, but also reasonably accurate.

Here's what it brings together under one roof:

  • Foundry Models: A catalogue of AI models from multiple providers - OpenAI (GPT-4o, o3), Anthropic (Claude), Meta (Llama), DeepSeek, Mistral, and others. You pick the model that fits your use case and deploy it through Azure.
  • Foundry Agent Service: A managed service for building and hosting AI agents that can orchestrate multi-step workflows, use tools, and maintain memory across conversations.
  • Foundry Tools: Azure's existing AI services - Speech, Translator, Document Intelligence, Computer Vision - repackaged and accessible through the same platform.
  • Foundry Control Plane: Governance, content safety, tracing, monitoring, and evaluation tools. This is where enterprise-readiness lives.
  • Foundry IQ: A knowledge integration layer that connects agents to enterprise data sources for grounded, citation-backed responses.
  • Foundry Local: Run models locally on your own hardware. Useful for development, testing, and scenarios where data can't leave the device.

The platform is free to use and explore. You only pay when you deploy models or consume services, which makes it straightforward to evaluate before committing budget.

For a full technical breakdown, see the official Microsoft Foundry documentation.

What's Changed from Azure AI Foundry

If you were using Azure AI Foundry (or Azure AI Studio before that), the rebrand to Microsoft Foundry isn't just cosmetic. There are real improvements:

The new portal is faster and better organised. There are now two portal versions - classic and new. The new portal is built specifically for agent development and multi-agent workflows. It's noticeably quicker to load and the navigation actually makes sense. The classic portal still exists for managing hub-based projects and Azure OpenAI resources directly.

Multi-agent orchestration is a first-class feature. The agent service now supports collaborative agent behaviour and workflow execution through SDKs for C# and Python. A real step up from the earlier single-agent patterns.

The tool catalogue has grown fast. Over 1,400 tools are now available, with both public and private catalogue support. For AI agent development, this means you can connect agents to enterprise systems without building every integration from scratch.

Memory and knowledge integration are built in. Agents can retain context across interactions, and Foundry IQ gives you a structured way to ground agent responses in your enterprise data. Both of these were painful to build manually before.

Where We've Seen It Work Well

We've deployed Microsoft Foundry across a range of projects for Australian businesses. Here's where it genuinely shines:

Model Flexibility

The model catalogue is the standout feature. GPT-4o, Claude, Llama, DeepSeek, and others through a single platform with unified governance. You pick the right model for each task without juggling multiple vendor relationships.

In practice, we use different models for different parts of the same system. A fast, cheap model for classification. A frontier model for complex reasoning. A specialised model for document processing. Foundry makes this practical rather than theoretical.

Enterprise Governance

For Australian organisations in regulated industries - financial services, healthcare, government - the governance story is strong. RBAC, network controls, content safety filters, audit logging, and Azure Policy integration all work through the same resource provider. Your existing Azure governance framework extends to AI workloads without a separate stack.

This matters because the biggest blocker to AI adoption in Australian enterprises isn't the technology. It's getting through internal security and compliance reviews. When AI runs inside the same governance boundary as everything else in Azure, those conversations get a lot simpler.

Agent Development

The Foundry Agent Service has matured a lot. Building agents that orchestrate tools, maintain state, and handle multi-turn conversations is much easier than it was twelve months ago. C# and Python SDKs both work well, so .NET shops and Python shops can build on it natively.

We've built agents for document processing pipelines, customer service workflows, and internal knowledge retrieval. The platform handles hosting, scaling, and monitoring, so we spend our time on business logic instead of infrastructure plumbing.

What's Still Rough

I'd be doing you a disservice if I didn't mention where the platform still has gaps.

The dual portal situation is confusing. A "classic" and "new" portal running side by side, with some features only in one or the other. Teams ramping up find this frustrating. Microsoft is clearly migrating toward the new portal, but the transition period is awkward.

Documentation hasn't caught up. Screenshots in the docs still reference Azure AI Foundry in many places. Naming is inconsistent across docs, portal, and SDKs. If you're setting up for the first time, expect to spend some time figuring out which docs actually apply to your setup.

JS/TS and Java SDKs are still in preview. If your backend is Node.js or Java, you're working with preview-quality SDK support. C# and Python SDKs are solid. The others aren't production-ready yet. For .NET-based projects, you're in good shape.

Regional availability in Australia is limited for some models. Not every model in the catalogue is available in the Australia East region. You may need to deploy certain models to US or European regions, which adds latency and potentially complicates data residency requirements.

Pricing can be hard to predict. You're assembling multiple services - model inference, agent hosting, storage, tool calls - so estimating monthly costs before deployment takes real effort. We always run a cost modelling exercise before committing to production.

Practical Advice for Getting Started

If you're considering Microsoft Foundry for your next AI project, here's what I'd recommend based on our experience:

1. Start with a specific use case, not the platform. Don't spin up Foundry to "explore AI." Pick a concrete problem - automating a document processing workflow, building an internal Q&A agent, improving a customer-facing process - and evaluate whether Foundry is the right fit for that problem.

2. Use the new portal for agent work. If you're building agents, go straight to the new portal experience. It's purpose-built for agent development and you'll waste time trying to do agent work in classic.

3. Test multiple models early. The model catalogue is Foundry's biggest advantage. Don't default to GPT-4o for everything. Run your actual prompts across 3-4 models and compare outputs, latency, and cost. You'll be surprised how much the best model varies by task.

4. Set up governance from day one. Don't bolt on RBAC, content safety, and monitoring after the fact. Set it up when you create your Foundry resource. It's far easier to start with the right controls than to retrofit them.

5. Plan for cost visibility. Set up Azure Cost Management tags and budgets early. AI costs can scale quickly once agents are in production, and you want visibility before you get the bill.

How This Fits with the Rest of the Microsoft AI Stack

Foundry doesn't exist in isolation. It sits alongside Copilot Studio (for no-code agent building), Microsoft Fabric (for data and analytics), and Power Platform. Understanding how these pieces connect matters.

For organisations that want business users building simple agents without code, Copilot Studio is still the right starting point. Foundry is where you go when you need custom logic, multi-agent orchestration, or integration with models beyond what Copilot Studio supports.

For data-heavy AI projects, connecting Foundry to Microsoft Fabric for data pipelines and storage gives you a solid end-to-end architecture. We've built several projects where Fabric handles the data preparation and Foundry handles the AI inference and agent logic.

The Bottom Line

Microsoft Foundry is the most complete AI development platform on Azure to date. The model catalogue gives you genuine choice. The agent service is production-ready for many use cases. The governance features meet enterprise requirements without bolting on third-party tools.

It's not perfect. The portal transition is messy, some SDKs are immature, and the rapid pace of change means what you learn today may shift in six months. But the direction is right, and for Australian businesses committed to the Microsoft ecosystem, it's the obvious place to build AI.

If you want help evaluating whether Microsoft Foundry is the right fit for your next AI project, or you need hands-on support getting from prototype to production, reach out to our team. We've been building on this platform since day one and we know where the sharp edges are.