From conversation to organisation-specific actions

Copilot Studio combined capabilities from Power Virtual Agents with generative AI and connections to Microsoft’s wider platform. A subject-matter expert could describe the intended assistant, connect approved data sources, define conversation topics and publish it to selected channels. Microsoft also presented a way to extend Microsoft 365 Copilot with an organisation’s own data, plugins and actions.

The availability status had two parts. Copilot Studio was made available as a product, while the ability to create extensions for Microsoft 365 Copilot was described as a preview. It was therefore too early to treat the entire announcement as one finished production package. Organisations still had to confirm which capabilities, regions and licence terms applied to their environment.

A fast-moving market

TechCrunch described Copilot Studio as Microsoft’s response to a growing market of tools that connected language models to proprietary business data. Computerworld quoted Gartner analyst Jason Wong on enterprise demand for extending Copilot beyond its standard capabilities. Both perspectives showed that the competition involved more than a capable chat interface: integration with data and systems was becoming decisive.

OpenAI had introduced function calling in June, enabling a model to return structured arguments for external functions. That development explained why actions and connectors became central to Copilot Studio. OpenAI’s own description also showed that the model proposed a function call; application code still had to validate and execute the action. That distinction mattered in low-code products as well.

Governance had to begin with the first pilot

The simplest starting point was a bounded knowledge assistant using a small set of known sources. Such a pilot could measure whether people found correct answers faster and whether responses pointed back to current content. Before an assistant could create cases, send messages or change data, its owners needed to define approvals, logging, and how an incorrect action would be stopped and corrected.

Access travelled with the content. An assistant that could retrieve CRM data or internal documents needed to observe the user’s existing permissions and avoid making restricted information easier to discover. Source quality, refresh frequency and ownership therefore became part of the product rather than a separate clean-up exercise.

The announcement made custom assistants accessible to teams beyond traditional software development. It lowered the barrier to building, but not accountability for the outcome. Management needed to choose a process with clear risk and value, require human control for consequential actions, and expand only after measurements demonstrated consistent quality.

The organisation also needed a simple record of unanswered questions, missing sources and rejected actions. This failure log provided a more realistic view of maturity than successful demonstrations alone and revealed where training or content work was still required.

Kilder

Microsoft: Introducing Copilot Studio, 15 November 2023

TechCrunch: Microsoft unveils Copilot Studio, 15 November 2023

Computerworld: Businesses can customise Microsoft’s assistant, 15 November 2023

OpenAI: Function calling and other API updates, 13 June 2023

Further reading

Read more: What is Copilot Studio?

Join the discussion

Which experience from your own organisation is most relevant to this story? Share what you learned.