If you’ve been tracking the surge in AI tooling lately, you’ve probably felt overwhelmed by the sheer pace of change. Build 2025 brought a raft of Copilot Studio announcements that promise to make agent-driven workflows more powerful—and more accessible—than ever before.

The new SDK

I’m particularly excited about the Microsoft 365 Agents SDK. Think of it not as a replacement for the Bot Framework, but as an evolution that brings “agents” to the centre stage. If you’ve built bots before, this SDK will feel familiar—yet it unlocks new patterns for composing and orchestrating intelligent assistants. If you want to learn more about agentic patterns, then please check out the excellent post by Jeff Williams found here.

Agent store and custom models

Have you ever wished you could share your custom agent with colleagues as easily as publishing an app? The Agent Store is now generally available, letting you package and distribute your work in a marketplace-style format. Even better, you can “bring your own model” from Azure AI Foundry—fine-tune it for your industry, tweak summarisation prompts, or swap out the default Copilot brains for your proprietary AI.

Multi-agent orchestration

What if a dozen specialised agents could coordinate to solve a complex task? Copilot Studio’s multi-agent support—in conjunction with the Model Context Protocol—lets you define parent-child relationships and negotiate duties across agents. I’d caution against over-componentising every tiny feature (that’s what component collections excel at), but for scenarios like cross-functional workflows or third-party integrations, this is a game-changer.

Automation tools Remember when RPA felt like magic?

The new “computer use” tool brings that magic into Copilot Studio: screen scraping, form-filling, web navigation—all without an API. It’s going to be huge for data-entry automation or for legacy systems where you simply can’t get a direct connector. And with the code interpreter now embedded, your agents can run Python on the fly—think data analysis, visualisation and complex calculations without leaving the platform.

Data and knowledge integrations

Copilot Studio already powers RAG (retrieval-augmented generation) patterns, but with Azure AI Search and Dynamics 365 data coming to M365 Copilot, the possibilities multiply. Whether you’re indexing ZenDesk tickets, ServiceNow records or your own SharePoint library via drag-and-drop, you’ll get fine-grain control over how knowledge is fetched and summarised.

Embedded builder enhancements

Several UX improvements in the embedded agent builder—everything from entity-knowledge components that tap your Teams chat history to direct file upload for ad-hoc knowledge. For anyone juggling multiple inboxes, calendars and documents, this feels like an overdue but very welcome upgrade.

The Bottom Line

Copilot Studio’s Build 2025 updates aren’t just incremental—they signal a maturing platform that spans code-first SDKs, low-code stores, multi-agent collaboration and end-to-end automation. If you’re exploring AI use cases in your organisation, now is the time to experiment: spin up an agent in Copilot Studio, connect it to your data sources, and see how much manual work you can offload. If you are looking for support to get started then we at Olympic33 have been helping companies just like your get started on this journey, please don't hesitate to get in touch.

‍