Microsoft Word’s spell check can already suggest words in real time, but imagine an “idea check” future in which AI suggests ideas on the fly as users type. This is how Microsoft is evolving Copilot — as an AI assistant predicting ideas in real time by tapping into unique corporate and user data. That’s an idea the company shared for Microsoft 365 as generative AI technology sinks deeper into the guts of the software. Specifically, Microsoft on Monday released a feature called Copilot Tuning, which will allow enterprises to deploy specialized AI models and agents that are fine-tuned to company operations and data. The feature was announced at Microsoft’s Build developer show being held this week. With Copilot Tuning, companies can train models on their own data and deploy AI agents that make use of their unique terms, context, and processes, said Jason Henderson, corporate vice president of Microsoft 365 product management, in an interview with Computerworld. “AI is going to have to … integrate the knowledge from your work into your specific Copilot so it can customize documents and workflows to the way your company works,” Henderson said. The larger goal behind Tuning is to create an “enterprise brain” that is a digital representation of an organization’s unique way of handling information, processing it and making decisions. That is key to creating Copilots that can recommend ideas, Henderson said. The ability to fine-tune AI to internal data changes the way model reasons and responds. That, in turn, changes how employees create documents and generate summaries. Henderson gave the example of a longstanding law firm with more than 100 years of institutional knowledge and precedents. The partners are retiring and taking their knowledge with them. “What they’re really interested in is fine-tuning into these models all these legal agreements that they have built over many, many years. And then being able to create new ones based on that learning,” Henderson said. Microsoft Copilot launched in 2023 providing generic responses from genAI large language models (LLMs), notably OpenAI’s GPT series. About 70% of Fortune 500 companies are now using Copilot in some form, according to Microsoft. Although LLM-based chatbots such as Anthropic’s Claude, Google’s Gemini and especially OpenAI’s ChatGPT remain popular, some companies are shifting to smaller reasoning models that are fine-tuned to domain data, finding the smaller models cheaper, faster, more customizable, and more secure. Microsoft says Copilot Tuning is the starting point to creating hyper-personalized features, such as suggesting ideas in real time as a person writes a document in Word. “I can think of this as a spell check of the future where it knows the words that your company uses. It’s going to predict based on what it knows about your company. It might be one thing in manufacturing versus in pharmaceuticals. It’s trying to help you get one step ahead by bringing to your fingertips what you need,” Henderson said. Security is an important part of Copilot Tuning, and the data is usually pulled and referenced from Microsoft SharePoint, which monitors access to files. Microsoft is creating a chain of custody and security from documents to the model. “We’re starting with SharePoint, but with connectors, we can bring in many, many different systems,” Henderson said. Google, Anthropic, and OpenAI all offer ways to integrate enterprise data into LLMs as well.  The new Copilot Tuning offering doesn’t require data scientists or coders, and it has a point-and-click interface within Copilot Studio. It can be used by regular employees who may think, “hey, I want to make my department’s life easier,” Henderson said. Microsoft is also creating “recipes” that are targeted at generating documents and summaries, or for data analysis and accounting tasks. “We’re going to identify which models best handle those recipes, doing a lot of experimentation to pick the right model for each particular recipe. Over the next six months, we’ll continue to build out more recipes as we identify durable problem spaces that companies are interested in addressing with Copilot Tuning,” Henderson said. The Copilot Tuning functionality will begin rolling out to Microsoft 365 Copilot customers next month as part of its “Early Adopter Program.” Microsoft’s Copilot Wave 2 spring release — which includes a new Copilot app with new search, summarization, and personalization functionality — is now generally available.