Microsoft Copilot has evolved rapidly over the past several months. If it’s been a while since you last explored it, now is a great time to take a fresh look at what it can do inside your Microsoft tenant today.
Microsoft Copilot turned two years old last November. If you haven’t seen it lately, you won’t recognize it. It has evolved from a gangly pre-adolescent, full of promise but awkward and inexperienced, into a functioning young adult that is beginning to deliver on the promises made about it.
Too eloquent? Maybe, but it is the truth. Read on to see what is new and why you owe yourself another round with Copilot at your side.
“Copilot is not worth the cost.”
“Copilot does not live up to its promises.”
“Copilot is no ChatGPT.”
Every week, I hear IT and business leaders say one or more of these phrases during AI strategy sessions. However, depending on how long it has been since they last tried Microsoft Copilot, I sometimes agree with them.
From Early Disappointment To Reconsideration
When first launched, Copilot (and its non-Microsoft counterparts) promised amazing features and capabilities. Trial licenses were acquired, pilot programs rolled out, and many found that the promises were not in line with the actual experience.
Copilot struggled to understand complex spreadsheets. PowerPoints were basic and did not use organizational templates. Its chat experience was not as robust and comprehensive as what users experienced in ChatGPT. Users quickly entered a trough of disillusionment, and many wrote the product off.
I cannot argue with any of these statements. I’m not going to analyze the “why” here and now, but I will say this:
If you are a “Microsoft shop” and you have either never used Copilot or not tried it within the last 2 months, you owe it to yourself, no, you owe it to your users and company, to try it again. (Bold and italics added by me! Not AI)
Why, you ask? Let me break it down into three parts…
1) Latest OpenAI Models
As of this writing, Microsoft Copilot uses the latest GPT-5 model by default. It can also switch to a Think Deeper mode that uses o-series reasoning when required. Users have the option to select GPT-5.2, which OpenAI describes as the most advanced frontier model for professional work and long-running agents.
The earlier versions of Copilot did not use these models and typically lagged behind in the models available for use. That has changed. While responses are still not perfectly aligned with ChatGPT due to orchestration differences, users now benefit from reasoning over their internal content with the latest models available.
Model Selection and Prompt Discipline
Show them how to use the dropdown menu in the upper right corner to choose the model they want. If a prompt does not work with one model, try another. If it still does not work, take a step back and re-evaluate your prompt.
Did you provide enough instruction? Did you supply context? Did you treat Copilot as you would a capable teammate entering a situation without your history and knowledge?
Approach it with that mindset the next time you are underwhelmed by the output. Spending a few extra minutes on clarity will pay off in quality.
2) Agent Mode
This is a feature I am genuinely excited about. Agent Mode in Excel, PowerPoint, and Word allows Copilot to directly manipulate document content based on your instructions.
This goes far beyond adjusting fonts or applying conditional formatting.
Real-World Example: Project Reallocation
Two weeks ago, I received the unfortunate news that one of our engineers would be moving on. We track projects in Azure DevOps, and I faced the task of reshuffling the schedule. Agent Mode had just appeared in our tenant, so I decided to put Copilot through its paces.
I performed a basic export of our project assignments and passed the delimited information into Copilot in Excel. I explained what had happened and what I needed done: a clean reallocation of projects to other engineers.
What happened next put a smile on my face.
Copilot understood the raw extract, even extracting dates embedded in sprint names. It created temporary tables, corrected its own logic assumptions, and asked follow-up questions.
I iterated with Copilot, adding PTO metrics, engineer expertise, and additional constraints. In the end, I had a clean schedule embedded in the same spreadsheet.
There is no way Copilot could have pulled this off a few months ago.
Also, I tried another use case where I asked Copilot to evaluate a set of Word-based use cases and place them into our evaluation matrix. Again, its ability to follow instructions was precise and aligned with expectations.
It is still not perfect. But we are much closer to the collaborative vision originally described by Microsoft.
Agent Mode is rolling out now and can be accessed by clicking the Tools icon within the Copilot prompt box.
3) Agents
Did you think I would get through this without mentioning agents? Here they are.
Copilot includes two core agents worth mentioning: Researcher and Analyst.
Put simply, a Copilot agent is a task-specific AI assistant embedded inside Microsoft 365 apps to perform structured work such as research, analysis, or document generation.
Researcher Agent
Researcher excels at detailed research that pulls in internal work data and web sources to create multi-page reports. These are not simple bullet lists. They are structured documents that read like professional reports.
You can iterate on the output and, once finalized, transfer it directly into destinations such as PowerPoint.
I have had significant success using Researcher to create first and second drafts, then layering in my own voice and style. With continued interaction, it increasingly adapts to writing patterns and tone.
Analyst Agent
Analyst focuses on data reasoning. It works across spreadsheets to extract trends, insights, and decision-support information.
Expanding the Agent Ecosystem
Microsoft has also introduced Word, Excel, and PowerPoint agents that expand how you generate new content using natural language.
Add custom agents built in Copilot Studio or Agent Builder, whose consumption is covered by the Copilot license, and the value multiplies.
There are also more than one hundred third-party agents available for deployment within your tenant.
Enterprise Protection and What Comes Next
All of these capabilities operate within Microsoft’s Enterprise Data Protection framework, meaning your data remains within your tenant and is not used to train foundation models. This security boundary is critical for organizations handling regulated or sensitive information.
Beyond these capabilities, there is also the Frontier program, an early release option that can be enabled for select users. It introduces advanced features such as coding agents, computer use capabilities, and Claude integration, positioning Copilot as a control layer across multiple large language models.
Copilot will continue to evolve.
Give your users the training and reinforcement they need to fully leverage these tools.
Need help sorting through these features and figuring out how to effectively re-engage? Let’s find time to connect and explore what a structured Copilot adoption strategy could look like for your organization.
Re-Engage Copilot with Structure, Not Guesswork
Copilot delivers real productivity gains when aligned to defined use cases, role-based enablement, and measurable outcomes. We help organizations move beyond pilots into structured deployment and adoption.
