Analysis
Most of Microsoft's AI playbook comes down to one idea: meet people where they already work. OpenAI builds the models; Microsoft wires them into Outlook, Teams, Word, and the rest of the software a billion-plus office workers open every morning. Copilot Studio, the low-code tool for building custom AI agents inside that world, is the clearest example yet.
The June 2026 release is the version worth paying attention to. What started as a way to spin up a basic bot now does enough that it lines up against the likes of OpenClaw, Anthropic's Dynamic Workflows, and Google's Agents CLI. (Those three aren't strict apples-to-apples competitors, mind you, Dynamic Workflows is a Claude Code feature rather than a standalone platform.) But Copilot Studio has a card none of them hold: it already lives inside the apps your team uses all day.
For an Australian business, the practical question isn't whether the technology is clever. It's whether you can hand it to a capable operations or finance person and get something useful out the other end without hiring engineers. That's the bet Microsoft is making, and it's the lens worth keeping in mind as we go through what the platform actually does.
The Platform Capabilities
The core idea behind Copilot Studio is letting non-developers build agents that work. You get a visual designer where you describe what the agent should do using plain-language instructions, some conditional logic, and ready-made action templates. Basic agents need no code at all. Power users who want more control can drop into Power Fx, a formula language built on the same logic as Excel formulas.
Microsoft groups agents along a couple of lines: conversational agents that answer questions and run tasks, and autonomous agents that sit in the background watching data and acting on it. The article framing here adds a third "hybrid" category that mixes both, though that three-way split is a useful description rather than Microsoft's official taxonomy. Either way, it covers most of what businesses ask for, from customer service to IT operations to sales support.
The June 2026 update brought a few changes worth calling out. Multi-agent orchestration lets several Copilot agents work together on a job, with one agent kicking off others when certain conditions are met. Custom tool integration through Power Platform connectors opens the door to more than 1,200 external services (Microsoft's own docs now cite over 1,400, so the figure is conservative). And the governance features give IT administrators a way to see and control every agent running across the organisation.
The Microsoft 365 Advantage
The integration with Microsoft 365 is where Copilot Studio pulls ahead. Agents built in it can read and act on data in Outlook, Teams, SharePoint, OneDrive, Excel, and Dynamics 365, with permissions handled through Microsoft Entra ID (the service formerly known as Azure Active Directory). That unlocks jobs that are awkward or near-impossible on a standalone platform:
- An agent that watches your Outlook inbox, spots emails that need a reply, drafts responses in your usual writing style, and hands them to you for approval
- An agent that reads Excel files in SharePoint, flags anomalies or trends, and writes up a summary in Word
- An agent that follows Teams conversations, pulls out action items, and creates tasks in Planner with the right people and deadlines attached
You could build all of this on other platforms using APIs and custom code. The difference is that Copilot Studio gives it to you ready to go, with security, compliance, and permission management that would otherwise take months to stand up.
Adoption and Revenue
Adoption numbers here come with a caveat. The figures originally circulated for this story put usage at over 350,000 agents since launch, with 120,000 in Q2 2026. Those numbers don't line up with anything Microsoft has said publicly. In its own earnings reporting, Microsoft claimed more than one million custom agents and 230,000-plus organisations using Copilot Studio, well above the smaller figures, so treat the 350,000/120,000 numbers as unconfirmed.
Pricing is where it pays to read the fine print. Standalone Copilot Studio starts at $200 per month for organisations without an eligible Microsoft subscription, though that's per 25,000-credit capacity pack on a tenant-wide licence rather than a flat per-user fee, with a pay-as-you-go option alongside it. Reports that the platform comes bundled into a $20/month Copilot Pro plan are out of date: Microsoft retired standalone Copilot Pro in late 2025, and Copilot Studio's internal use now rides with the Microsoft 365 Copilot add-on (around $30 per user per month). Claims that it's "effectively free" through Microsoft 365 E5 don't hold up either, Copilot is an add-on, not part of E5.
On revenue, the picture is fuzzy because Copilot Studio is bundled with other Microsoft AI products. One estimate put the whole Copilot family, Studio, M365 Copilot, and GitHub Copilot, at $5-6 billion in annual revenue, but that sits at the high end and isn't clearly sourced; other analysts land closer to $2.5-3.5 billion after enterprise discounting. What's not in doubt is that Microsoft's broader AI business is running at roughly a $37 billion annual rate.
Limitations
The trade-offs follow from who the platform is for. Because it's built for non-developers, it doesn't give engineers the flexibility they'd want for genuinely complex applications. Model choice has historically been limited to what Microsoft ships through its OpenAI partnership, though by 2026 the platform has added model-choice and bring-your-own-model options, so the old "you get one model, take it or leave it" framing no longer quite fits. Anything that needs to reach well outside the Microsoft ecosystem still calls for workarounds. And the visual designer, easy as it is to start with, gets unwieldy once an agent's logic grows.