First Encounters with Microsoft’s Copilot Researcher & Analyst

On Week 2 of #SummerOfCopilot (which started as #MonthOfCopilot but I had too much to say!), we dove deep into Microsoft's two newest AI agents: Researcher, your digital research assistant, and Analyst, your on-demand data expert. These innovations promise to change the way we approach research and analytics — moving Copilot from simple Q&A into genuine reasoning and analysis.

We also looked at Microsoft’s new “Copilot Tuning” which will let you recalibrate how Copilot answers certain types of queries and where it goes to resolve them.

Here’s our full recap and first-hand impressions.

Researcher: Qualitative Brilliance on Call

Announced in 2025, Researcher is designed to tackle complex research tasks that used to require hours of manual digging. By leveraging both your organization’s internal data and trusted web sources, Researcher can analyze, summarize, and deliver actionable insights.

We tried drafting market trend forecasts for the generative AI sector. What used to take a full afternoon is now a matter of minutes. The system’s multi-step query handling and synthesis elevate it beyond a typical search tool, pushing Copilot into true analytical territory. As it's working, the agent shows you what it's doing, and you can expand the full history along the way.

Researcher even looks past my lousy typing

I was able to quickly gather a table of different research forecasts for different years; this allowed me to apply my own weights and integrate our own research to create a new composite forecast.

(By the way our forecast will be shared later this summer. No surprise, the AI market is large and growing!)

Since its pre-release as part of the Frontier program, Researcher has made notable strides in speed, reliability, and the sophistication of its results. Its ability to capture and synthesize initial sets of information, especially for tasks like compiling market trends or drafting overviews—has become a real asset in daily workflows.

For example, when tasked with building a research report on three trends in the produce market, Researcher quickly assembled a credible and diverse list, such as apples, bananas, and carrots, pulling from both internal and trusted external data sources. Where Researcher truly excels is in the initial stages of information gathering -- it’s fast, context-aware, and generally surfaces relevant results that would otherwise take hours to assemble manually. However, when it comes to iterative refinement—such as requesting that “carrots” be swapped for “corn” or asking for the same information in several different formats over multiple chats — Researcher can become unwieldy.

In these scenarios, it tends to revert to the original dataset or regenerate earlier versions rather than persistently tracking your requested changes. “Carrots” kept coming back. This can make detailed message polishing or custom reformatting a cumbersome process, particularly for complex or evolving projects.

But that’s ok! Researcher was designed for research, not text reformatting. There are better tools for the job.

Our recommendation: Leverage Researcher for its strengths in initial research and rapid synthesis, but once you move to the stage where nuanced editing, reformatting, or repeated adjustments are needed, transition to regular Copilot. Copilot's flexibility and responsiveness make it better suited for the iterative polish required to produce high-quality final outputs.

What could be next for Researcher

The future presents additional opportunities to enhance Researcher capabilities, particularly with the new Copilot Notebooks feature.  Researcher could autonomously update its research with new articles, integrate these new sources, and notify you of its updated findings.

In July, Microsoft plans to add one-click generation of briefing documents and FAQs to Copilot Notebooks. This feature aims to streamline the documentation process, making it easier for researchers to compile and share their findings with stakeholders.  One click generating is great – and image Researcher being able to build those elements automatically (“zero-click”).

Analyst: Quantitative Insights, Explained

Analyst is the data scientist you didn’t know you needed. It securely connects to your data sources, parses scattered datasets, and generates not just answers but also clear visualizations and code. One standout: you can see and audit the Python code used in any advanced analysis. For teams without full-time analysts, this is a game-changer—especially when you need to spot patterns or anomalies hidden in messy real-world data.

First Impressions: Groundbreaking Yet Early

Both agents left us excited for the future. Researcher’s ability to deliver synthesized, cross-channel answers is a clear productivity win. Analyst’s transparency and narrative approach to data make it accessible, even for non-technical users. However, during our tests, we ran into usage caps and found that nuanced, high-stakes analysis still benefits from human oversight. Language support is evolving (Analyst is English-first), and sometimes the tools simplify or miss subtleties.

The Road Ahead

Microsoft’s cautious rollout, featuring usage limits and a focus on feedback, shows they’re serious about refinement. These agents are now available to Copilot customers. We plan to continue testing, pushing their limits on real work scenarios, and sharing what we learn.

Bottom line: Researcher and Analyst are the beginning of a new era—AI agents as true co-workers. The vision is here; with time and iteration, execution will catch up.

Copilot Tuning and Early Use Cases

A Glimpse into the Future of Custom Enterprise AI

Microsoft’s Copilot Tuning marks a significant leap for organizations seeking to customize AI without technical barriers. By enabling low-code fine-tuning of models using internal files and workflows, Copilot Tuning empowers teams to shape AI tools that understand their unique language and processes. With built-in “recipes” for expert Q&A, document summarization, and content generation, businesses can now create intelligent agents trained on their own data—all while maintaining data security and respecting permissions.

Early adopters are already seeing the value. Pilot programs show how even minimal labeling can yield highly effective, specialized AI assistants, from drafting legal documents in a firm’s house style to answering niche technical questions in a consultancy. While the feature is still in preview and results may vary, the direction is clear: Copilot Tuning could soon democratize enterprise AI, letting every organization craft its own highly tailored Copilot that truly speaks its language and meets its needs.

Stay tuned for our ongoing reviews and best practices as these tools become part of everyday work.