David Fortin slogged through more than 100 Microsoft 365 Copilot agents so you don’t have to. The former Microsoft MVP turned independent consultant tested agent after agent, separating the transformative from the trivial. His verdict, summarized by Geeky Gadgets, cuts through the AI hype: just five agents deliver real workplace value today. Meet Researcher, Facilitator, Knowledge Hub, Excel, and Co-work. These aren’t experimental toys. They plug directly into the apps millions of workers already use, automating drudgery and surfacing intelligence without forcing anyone to learn a new tool.

If you’ve felt agent fatigue creeping in—another week, another “Copilot for X” announcement—this list is your lifeline. Fortin’s practical filter strips away agents that require exotic data setups, demand bespoke training, or only shine in demos. The winners solve concrete problems: hunting down information scattered across teams, turning chaotic meetings into structured work, taming sprawling SharePoint sites, wrangling Excel models, and coordinating cross-team collaboration.

Why Most Copilot Agents Fall Short

Copilot agents promise to be the next evolution of AI at work, moving from simple chat assistants to autonomous entities that can reason across data, take actions, and integrate deeply with Microsoft 365. The reality is messier. Many agents suffer from a narrow scope: they work brilliantly in a controlled sandbox but stumble when confronted with real-world messiness—legacy file structures, inconsistent naming conventions, permissions spaghetti. Others demand a level of data hygiene that simply doesn’t exist in most organizations.

Fortin’s testing was ruthless. He evaluated each agent on three criteria: immediate usefulness without custom development, reliability across common tasks, and seamless integration with existing workflows. Agents that required IT to reconfigure SharePoint or retrain entire departments didn’t make the cut. The five that survived are ready to deploy now, with minimal setup, and they pay back the effort within days.

1. Researcher: The Answer Engine That Reads Your Organization

Researcher acts like a corporate search brain that actually understands context. It doesn’t just match keywords; it reasons over the entire corpus of documents, emails, chats, and meeting transcripts you have access to. Ask it a complex question—say, “What was the Q3 decision on product launch timing, who opposed it, and what data did they cite?”—and it scours Teams messages, PowerPoint decks, and Excel sheets to assemble a coherent answer with citations.

Fortin found Researcher indispensable because it eliminates the “who has the file?” problem that plagues large projects. Instead of pinging five colleagues on Teams and waiting hours for replies, you get answers in seconds. It handles follow-up questions naturally, so you can drill down: “Show me the specific revenue projection slide from the exec review.” The agent navigates the graph of your Microsoft 365 data, understanding that a chat thread about the launch may reference a file that was later updated, and it surfaces the right version.

Under the hood, Researcher combines semantic indexing, Microsoft Graph, and the Copilot orchestration engine. It respects existing security boundaries—users see only what they have permission to access. That governance model solved one of Fortin’s major concerns: agents that accidentally expose confidential data. Researcher simply won’t show you something your user profile can’t already open.

2. Facilitator: Your Meeting Assistant That Takes Action

Meetings eat time, but they also generate the raw material for decisions. Facilitator captures that raw material and transforms it into action. During a Teams call, it takes real-time notes, identifies decisions, assigns action items, and drafts a summary that goes far beyond a simple transcript. The magic is in its follow-through: after the meeting, it posts the summary into the team’s channel, updates any related Planner tasks, and even sends personalized nudges to action owners.

Fortin highlighted Facilitator as the agent that makes hybrid meetings bearable. Instead of one poor soul frantically typing notes while trying to participate, the agent handles documentation automatically. More importantly, it bridges the gap between discussion and execution. “We’ll circle back on that” stops being a euphemism for “forgotten forever.” The agent remembers and reminds.

Testing revealed that Facilitator shines when you let it run in the background of recurring meetings. Over a quarter, it builds a searchable archive of decision history that Researcher can later mine. That turns your meeting history into an organizational memory, not a black hole. Of course, transparency is critical. Fortin recommends always informing participants that the agent is active and what data it captures. Microsoft provides tools to label transcripts and summaries clearly, so no one is surprised.

3. Knowledge Hub: SharePoint’s AI Brain Transplant

SharePoint sites often become digital dumping grounds. Without rigorous governance, they fill with outdated files, multiple versions of the same document, and content that no one knows exists. Knowledge Hub is an agent that sits on top of a SharePoint site collection—and, optionally, connected Teams channels—to automatically organize, curate, and surface knowledge.

Fortin called this agent a “governance ninja.” It can detect duplicate content, suggest archival of stale documents, and even generate human-readable summaries for each folder or library. New team members no longer need a tutor to understand the site’s landscape; they can chat with Knowledge Hub directly: “Where do I find the latest brand guidelines?” The agent responds with a link and a snippet, and if appropriate, a note that the document’s owner last updated it two weeks ago.

One feature that stood out in testing was its ability to enforce lightweight compliance without heavy-handed IT policies. For example, the agent can flag files that lack certain metadata or that haven’t been accessed in 18 months, gently prompting site owners to take action. This self-healing approach addresses the root cause of SharePoint sprawl: nobody likes being the digital janitor. Knowledge Hub automates the janitorial work and lets humans focus on creating value.

For enterprises, the governance implications are significant. It reduces the risk of data leakage because it systematically classifies sensitive information and can alert security teams when, say, a finance spreadsheet lands in a marketing site. Importantly, it respects site collection boundaries and permissions, so it doesn’t become a panopticon—it works within your existing security model.

4. Excel: Copilot’s Data Engine Gets an Agent Upgrade

Most Excel users barely scratch the surface of what the program can do. Copilot in Excel already helps with formula writing and chart creation, but the new Excel agent described by Fortin goes further: it treats the entire workbook as a living data model, capable of not just answering questions but also suggesting analyses, cleaning data, and even building lightweight predictive models.

You can converse with your spreadsheet in natural language. “Which product category had the highest margin growth in the Northeast last quarter?” The agent dynamically selects the right pivot, applies filters, and creates a chart, all while showing its work so you can trust the result. Need to combine data from three sheets and a live Power Query feed? The agent writes and executes the necessary M code behind the scenes, sparing you the archaeology of learning Power Query syntax.

Fortin cautioned that the Excel agent requires a certain level of data structure—raw, inconsistent data still confuses it. But for businesses that already use Excel as a frontline analytics tool, the agent reduces the barrier to advanced analysis. It also integrates with Power Platform, so an insight found in Excel can immediately trigger a Teams notification or an update to a Power BI dashboard.

The most compelling use case from the testing: the agent as a “data butler” for managers who don’t live in Excel. A sales director can ask, “What’s the commission implication if I give the team a 5% uplift this month?” The agent pulls the latest data, models the change, and presents a clear answer, all while documenting the assumptions. This turns Excel from a solitary number-cruncher into a collaborative decision-support tool.

5. Co-work: The Collaboration Co-pilot for Cross-Team Projects

Projects that span departments often sink under the weight of misaligned priorities, duplicate work, and invisible progress. Co-work is an agent designed to orchestrate collaboration in Microsoft Teams. It connects across channels, Planner, Loop components, and even third-party integrations via connectors to provide a single pane of glass for project status and coordination.

Fortin valued Co-work most for its ability to maintain context. In a typical project, someone posts a status update in a Teams chat, another sends an email, and a third updates a Planner task—and nothing links together. Co-work stitches these fragmented signals into a coherent timeline. It can answer questions like “Who is waiting on legal approval for the website update?” or “What are the three blockers for the launch campaign?” and provide a summary drawn from the team’s actual work artifacts.

The agent also simplifies accountability. When a task is due, Co-work pings the responsible person and offers to schedule a brief check-in if the task is at risk. It respects working hours and personal focus time, so it doesn’t become a source of notification fatigue. During the test, teams using Co-work reported a noticeable drop in the number of stand-up meetings, because the agent provided a real-time status view that replaced the need for synchronous check-ins.

Integration with Knowledge Hub and Researcher creates a virtuous cycle: project learnings become part of the organizational knowledge base automatically. That means the next project doesn’t start from scratch but can learn from past failures and successes, preserved in the system by the agents themselves.

Real-World Governance and Security Considerations

Deploying agents at scale raises legitimate concerns. Fortin’s testing surfaced three governance principles that organizations should adopt before rolling out these agents widely.

First, access control is paramount. All the recommended agents operate strictly within the user’s existing Microsoft 365 permissions. Fortin advises administrators to audit their SharePoint and Teams permissions before enabling agents—otherwise, the agents will faithfully surface data that was technically accessible but perhaps not intended for broad consumption. An upfront permission cleanup is a small investment for long-term security.

Second, transparency must be built in. Employees should always know when they’re interacting with an agent or when an agent is present in a meeting. Microsoft provides audit logs and compliance tools, but it’s up to organizations to educate users about what the agents can do and how their data is used. Fortin recommends a clear internal communication plan, perhaps even a short training video, to set expectations.

Third, start with a pilot group. The five agents here are stable, but every organization is unique. Pick a team that’s eager to experiment and let them use the agents for a quarter. Collect feedback, adjust, and then expand. Rushing a broad deployment risks backlash and the kind of “AI fatigue” that kills adoption.

The Road Ahead for Copilot Agents

Fortin’s shortlist is a snapshot of maturity, not a permanent ranking. Microsoft’s agent ecosystem is still young. As the underlying models improve—especially in reasoning and planning—the barrier to building custom agents will drop. Power Platform already allows low-code agent creation, and the Copilot Studio lets organizations define custom skills and knowledge sources.

But the real shift will come when agents start collaborating with each other. Imagine Researcher passing a data set to Excel, which summarizes it and hands the insight to Facilitator for inclusion in a weekly status meeting, all without a human orchestrating the steps. That level of multi-agent orchestration is on the roadmap, and the five agents here form the foundational building blocks.

Microsoft’s challenge—and opportunity—is to make this orchestration safe and explainable. No one wants a self-running AI that makes decisions they can’t trace. The agents identified by Fortin already keep humans in the loop by default, and that’s likely to remain a design principle. The future isn’t about removing people; it’s about removing the tedious connective tissue between the work that matters.

How to Get Started Today

Fortin’s advice for organizations ready to move: pick just one agent and deploy it where the pain is worst. If your team wastes hours searching for information, start with Researcher. If meetings are a black hole, begin with Facilitator. If SharePoint governance is a growing nightmare, Knowledge Hub will quickly prove its worth. Excel and Co-work address more specialized needs but can deliver transformative returns in the right context.

All five agents are available within Microsoft 365 Copilot subscriptions, with some features requiring specific license tiers. Fortin’s detailed walkthrough, available on his blog and summarized by Geeky Gadgets, provides step-by-step setup guides that avoid common pitfalls. The key takeaway from his marathon test: the era of useful AI agents at work has arrived, but it’s the practical, deeply integrated ones—not the flashy demos—that will change how people work.