If your business runs on Microsoft 365, you’ve already encountered Copilot somewhere.
Copilot is what powers that button that pops up offering to revise the email you’re writing and that AI-powered meeting summary that shows up when you end a Teams call.
We’ve written about a few features here and there within Copilot and covered a few new feature releases. Now that Microsoft has dropped its Wave 3 update to Copilot, there’s a lot more to discuss.
What Is Copilot? A Quick Refresher
Copilot is Microsoft’s brand of modern generative AI tools, encompassing every user-facing AI feature within Windows 11 and Microsoft 365. Users can ask Copilot questions, use it to refine their creative output, generate meeting summaries, and so on.
Admittedly, Microsoft’s move to Copilot hasn’t been controversy-free. Some users don’t love the very visible way Microsoft is rolling it out. Some businesses have privacy concerns, and some individuals and businesses alike don’t appreciate the non-optionality of it all: you can’t really turn most of it off even if you want to.
That said, existing Copilot capabilities are genuinely useful in lots of business situations. And if you’re using it actively so far, you’re probably using it to do the kinds of things we’ve already described — drafting, revising, summarizing, and question-answering.
But the Wave 3 update is pushing Copilot in bold new directions.
Wave 3 Signals a Shift
Most existing Copilot use cases help individual users do things, but the user is still the one directing the whole process. Copilot can draft an email, but only if you tell it what to say. It can refine your email, but only after you draft it. It can summarize meeting notes, but only when you tell it to and give it access.
The new capabilities in Wave 3 look and feel different.
You may have heard about agentic AI already: the idea where instances of AI (often called AI agents) do work with some level of independence. That’s what Wave 3 unlocks.
Now it’s possible to build small AI assistants that run continuously and take actions you define on an ongoing basis.
Instead of “hey Copilot, can you make this email I wrote sound more professional?”, now you might design an agent that can run a multi-step workflow on its own. An automated invoice processing agent might take these steps:
- Triggers when a new PDF lands in a designated inbox or folder
- Uses AI to pull out vendor name, amounts, items and so on
- Searches the CRM to verify PO exists
- If it does, routes the invoice for standard processing
- If it doesn’t, sends the invoice to a human for evaluation
- Updates a database or spreadsheet to record the transaction
And that’s just one of thousands of examples, many of which come prebuilt and ready for users to configure.
Isn’t That Just Business Process Automation?
Good question. Savvy readers might identify that several of those steps don’t necessarily use AI. And it’s true: Plain ol’ business process automation (BPA) can do a number of those steps.
The difference AI makes is in intelligently pulling it all together and dealing with uncertainty and unstructured data. AI can reliably pull information from unstructured data in a way that conventional BPA can’t.
It’s also becoming a whole lot easier to create an AI agent that can do something like that workflow, lowering the barrier to entry compared to conventional BPA.
Greater Capabilities, Greater Risk
Of course, the more capable and intertwined AI agents become — and the more access they gain to your data and systems — the greater the risk. An AI agent can deliver huge efficiency gains, but a misconfigured or “rogue” AI agent can create equally huge operational problems.
The right approach requires balance and, for many businesses, a helping hand from an IT service provider like us. We can help your team grow safely in the age of AI. Give us a call or schedule a consult today.