This story is featured in the WorkLab newsletter. Sign up for it here.

Over the past year, Microsoft researchers have been studying the effects of AI at work through a series of real-world experiments at companies that are using Copilot. The setup was simple: one group used Copilot, the other didn’t.  
 
Across the board, people using Copilot worked faster and produced better results. But our findings also surfaced new patterns in how people work, learn, and adapt. Here are four takeaways that leaders can apply to their own AI adoption as we move into an era of intelligence on tap and increasing collaboration between humans and agents. 

1. AI is already embedded in how we work 
In many of our studies, members of the control group, who were not supposed to use AI, used it anyway. This is consistent with our data showing that employees who are not provided with AI at work will seek it out themselves (BYOAI). There’s no such thing as a “non-user” anymore: AI is already embedded in how people think, work, and get things done. 

This presents an opportunity for every leader and company. Imagine you could go back to the early days of the internet, when individuals were already using it widely but companies were still figuring out how to apply it to business. What would you do differently? Frontier Firms—structured around on-demand intelligence and powered by “hybrid” teams of humans + agents—hold key lessons for how to grab this once-in-a-generation opportunity and are poised to get unprecedented value from AI. To seize the moment, start by hiring your first digital employees, setting your unique human-agent ratio, and driving broad, purposeful adoption. 

2. Scale what works 
To evaluate how well employees performed on complex, job-specific tasks, our researchers needed a way to assess quality—even though they weren’t experts in, say, root cause analysis. So they turned to Copilot, building an AI grader to help understand the quality of the responses they got. First, a leader at the company where the study took place validated what a high-quality version of the finished product looked like. Against that measuring stick, Copilot was able to objectively evaluate all study participants’ work and assign each a qualitative score. 

This is a powerful AI use case for business leaders. According to our latest Work Trend Index, 55% of employees at Frontier Firms say they’re able to take on more work—partly because they use Copilot to pressure-test deliverables, benchmark quality, and surface blind spots without waiting for manager review. This can be as simple as training an agent on how you and other leaders have critiqued your team’s previous work so employees can run a first-pass review. 

The overall takeaway: real AI transformation happens when organizations capture what works and scale it across teams. By formalizing best practices—as in the agent example above—and codifying them with AI, companies can ensure consistent quality and faster decision-making. This ability to operationalize AI learnings is what sets Frontier Firms apart from those still experimenting. 

3. Skilling can’t be an afterthought 
One theme that showed up across the Copilot research: the biggest performance gains actually came when study subjects received guidance on how to apply AI to specific job tasks. 

In one study, our researchers developed optimal prompts that subjects could use. Employees provided with these prompts and tips on how best to leverage AI saw the biggest positive impact on performance. The key is making the connection between what someone needs to do and how AI can help them do it better. 
 
To see meaningful results, leaders need to go beyond access and focus on enablement: clear guidance, real-world use cases, and training that connects AI directly to the work employees are doing. When people understand how to use AI with purpose, they’ll see better results. 

4. Get your processes in order 
Just as targeted guidance helps individuals, structured processes help teams get the most from AI. Across our experiments, the biggest improvements came when participants had clear goals, well-defined tasks, and access to organized resources—like structured SharePoint libraries or shared templates. In those cases, Copilot acted as a powerful amplifier. But when processes were disorganized—unclear roles, messy documents, lack of shared understanding—performance improvements were limited.  

AI can accelerate work, but it can’t untangle dysfunction. Before layering on AI, clarify goals, align teams, and clean up the systems people rely on every day.  

When it comes to AI transformation, we’re all learning as we go. Becoming a Frontier Firm is about experimentation, capturing what you learn, and quickly applying those lessons to scale fast—empowering everyone for a new world of work.