1300 166 138 Tell us what you need
News / Microsoft

Microsoft opens up the engine behind Copilot to Aussie firms

By Monika Tantau • Published 2 weeks ago
A business professional in a blue blazer studies his computer screen with a thoughtful look, weighing up what Microsoft's Copilot harness could mean for his company

Key takeaways

• Microsoft released the Copilot harness, the layer that runs Copilot's agents, through Microsoft Foundry on 28 October
• It works with more than 20 AI models, plus your own API keys
• Microsoft runs the agent loop and the sandbox; you bring the instructions, tools and model choice
• Pricing and regions are not published yet, so Australian teams should ask before building on it

Most Read

Microsoft has opened up the engine that runs its own Copilot products. The Copilot harness, built by GitHub’s engineers, now powers Copilot Studio, the GitHub Copilot app, VS Code and more, and from 28 October businesses can build their own agents on it through Microsoft Foundry, according to Microsoft’s announcement.

For the many Australian organisations already paying for Microsoft 365 and Copilot, this changes a familiar question. Instead of building and maintaining the plumbing around an AI agent themselves, they can rent the same plumbing Microsoft uses.

What a harness actually is

An AI model supplies the intelligence, but on its own it cannot do much work. The harness is everything around it: the instructions, the context, the tools the agent can call, the place where its code runs and the controls that stop it going off track. Microsoft describes it as the orchestration layer that lets an agent reason, use tools and finish long, complex tasks.

Building one is harder than it looks. A harness has to stay fast and cheap, keep working as new models arrive every few weeks, and support open standards such as MCP, agent skills and plugins. Microsoft’s pitch is that most companies should not have to maintain that layer themselves.

What Microsoft is offering

The harness works with more than 20 frontier models, and customers can also bring their own API keys. Microsoft says it can route each job to the model that suits it, so a team does not have to test every new release before switching.

There are two ways in. Developers can download an SDK, or run the harness in the cloud with Foundry managing it. In the managed version, Microsoft runs the agent loop and the sandbox with governance built in, while the customer supplies the instructions, skills, tools and choice of model. Microsoft also says Foundry stays open to other harnesses and frameworks.

Hands holding a tablet showing the Microsoft Copilot start screen with Word, Excel, PowerPoint, Outlook and OneNote icons, the products the Copilot harness already powers

Why Microsoft says it can be trusted

The argument is scale. The same harness runs across Copilot surfaces used by millions of people, and Microsoft says it recently rewrote the runtime in Rust, cutting the overhead of each session and the wait for the first response. Every improvement it makes for its own products reaches customer agents built on it too.

That is Microsoft’s claim, not an independent test. The announcement is written by Tina Schuchman, who runs Foundry, and Vladimir Fedorov, GitHub’s chief technology officer, and it does not include benchmarks against other agent frameworks.

Why Australian businesses should care

Many Australian companies run on Microsoft already, so the harness lowers the effort of moving from a Copilot chat to an agent that completes a real process, such as triaging support tickets or preparing reports. Teams weighing it against tools like Claude Code and Codex now have a managed option from the vendor they already pay.

The missing details matter here. Microsoft has not published pricing, plan tiers or which Azure regions the managed harness runs in. Before building on it, ask Microsoft whether agents and their data stay in Australian data centres, which matters under the Privacy Act and for sectors with residency rules. Our guide to AI data residency covers what to check.

A sign of where agents are heading

The bigger shift is that the harness is becoming a product in its own right, separate from the model. Microsoft is betting that companies will switch models often but keep the same engine around them, and that the engine is where vendors can lock in customers.

Choosing which layer to own and which to rent is now part of any AI strategy. The next things to watch are the price list and the first customer agents built on it outside Microsoft’s own products.

Scroll to Top