AI at Work, One Year Later: Copilot, Fabric, and What’s Actually Changed

Introduction

Last year, I wrote about how Microsoft Copilot changed how I work in “Unlocking Productivity with Microsoft Copilot: A Year in Review.” This year the story is bigger than Copilot. I’m seeing the best results when AI is integrated into the systems we already live in such as Microsoft 365 for day-to-day work, Power BI/Fabric for a “single pane of glass” for operational data, and a few other tools that help with both coding and learning.

The other reason I’m paying attention, I recently listened to students describe co-op workplaces that were AI-heavy, AI-available, and no-AI—and the gap was real. In the no-AI environment, the student was doing work they knew could be automated. That stuck with me because it’s not just a student issue, it’s potentially a career issue for all of us as well as for companies that are trying to stay competitive.

Copilot: Continual Evolution

A few updates have stood out in day-to-day use:

• Model upgrade: In the M365 Copilot tool it now shows GPT‑5 as the large language model (LLM) being used. I’ve found that this is now bringing me better output quality then the previous LLM, especially for longer-form writing, summarization, and “review/critique” style requests.
• Fast vs. deep responses: This choice is now an option within M365 Copilot’s user interface. Depending on what I’m doing, I’ll either ask for a quick answer or a more thorough analysis (like reviewing a draft blog post). Copilot has gotten noticeably better at both.
• New Library tool in the M365 Copilot app: There’s now a Library that makes it easier to find and reuse Pages & Images created with Copilot.
• Workflows in Word and Excel: Copilot appears to support more step-by-step workflows, apply changes in the document/workbook, and let you steer revisions as it goes. This is not a feature I’ve leveraged yet, but I can see how it would be useful for the right business case.

Microsoft Fabric: Results From Our Test Drive

If you’re not familiar with Microsoft Fabric, think of it as Microsoft’s platform for bringing data, analytics, and reporting together in one ecosystem. We approached it the same way we approach most new platforms: start small, test with limited workloads, and see whether the value shows up in day-to-day use.

The problem we were solving

We have an extensive data warehouse pulling from 6+ different data sources tied to our ticketing platform, CRM, plus data from our customers. Individually, each system tells part of the story. The time sink is switching between tools, reconciling what’s “true,” and building a coherent timeline of what changed and when.

What we built: a “single pane of glass” in Power BI

By consolidating these sources into Power BI, we were able to give users a single view of client and operational data across systems. The immediate payoff was simple: fewer context switches and far less time spent hunting for related details spread across multiple tools.

Where AI-style features added leverage: Narrative Cards + Copilot Q&A

Once the cross-system view was in place, we added two AI tools on top: Narrative Visualizations and Copilot in Power BI. The goal for both: get people to “what matters” faster, without having to navigate data in multiple systems.

Narrative Visualizations are prompt-driven, dynamically generated summaries that turn report context into readable, decision-oriented language.

Before we added Power BI Narrative Cards, getting fully oriented on a client often meant an hour of reviewing tickets, time entries, and agreements to try to stitch together a coherent picture. Now, with that data unified and summarized, the “what’s going on here?” question is now answered in seconds.

A couple prompts/use cases that have been especially effective:

• “What changed in this client’s environment in the last 24 hours?”
  Instead of checking multiple dashboards and logs, the Narrative Card summarizes key data from multiple relevant sources.
• “What’s the most important context everyone should know about this client right now?”
  This is useful for prepping for a call, doing a handoff, or just getting your brain back into context fast. It includes information on priority tickets and ongoing projects.

On top of that, Copilot in Power BI lets users ask plain-English questions against that same data. When it works well it gives people a shortcut to “query the warehouse” without needing to know where the data lives or how it’s joined.

The impact isn’t that any of these replaces the underlying metrics. It’s that it surfaces what matters first, and then you can drill into the visuals and source details when needed.

Why this matters

For us, the real win is the combination:

• Centralized reporting across many operational systems (the “single pane of glass”), plus
• Prompt-driven summaries and natural language interaction that reduce time-to-understanding.

It’s not magic, and it doesn’t eliminate the need for good modeling or governance, but it does change how quickly people can go from “something happened” to “here’s what changed, here’s why it matters, and here’s where to look next.”

Exploring Other AI Tools

While Copilot is the most visible AI tool in the Microsoft ecosystem, we’ve also been exploring tools outside of Copilot—especially where the use case is more service desk, customer support, and knowledge-driven automation than document-centric productivity.

AI assistance for ticketing and customer support

One tool we’ve been evaluating connects into our ticketing workflow and helps reduce friction for both customers and our team. The most promising areas so far:

• Ticket classification and routing: Helping categorize incoming tickets more consistently, improving triage and reducing time-to-assignment.
• Drafting responses: Generating first pass replies that technicians can review and tailor, useful for common questions and repeatable issues.
• Customer self-service ticket creation: Guiding customers to submit better tickets (more complete details, clearer symptoms), which speeds resolution.
• Customer-aware chatbot support: Answering basic questions via chat using context tied to the specific customer—based on what we already know about their environment and history.

The goal here isn’t “replace support.” It’s to reduce repetitive work so the team can focus on higher-value troubleshooting and better customer outcomes.

AI platform to help customers adopt AI

Systems Engineering is evaluating a platform that could help customers adopt multiple large language models (LLM) in a practical and controlled way.

The reason it’s compelling is flexibility. Different customers have different needs, and we see the need for tools that provide a way to leverage LLMS without forcing everything into a single “one size fits all” tool. It’s a platform approach that can support everything from quick productivity gains to more repeatable, packaged workflows.

A simple example from my own work: for this blog post, I dictated several brain dumps of notes, brought it into a tool with the LLM option I wanted, and used it to turn raw thoughts into a clean blog-post structure. That saved time and produced a stronger first draft than I would have written from scratch. I still did a manual editing pass for accuracy, tone, and to make sure it reflects what happened.

The bigger point is the same thing we tell customers: AI works best when it’s used intentionally, with the right guardrails, and with a human still responsible for the final output.

Why we’re looking beyond one tool

Different tools are optimized for different workflows:

• Copilot is great when work lives inside Microsoft 365 (docs, email, spreadsheets, slides).
• AI assistants are compelling when work starts with tickets, customer questions, and operational knowledge.
• Multi-LLM Platforms can help when you want flexibility across models and repeatable workflows that can be packaged for customers.

The bigger takeaway: the AI landscape is widening, and the “best” tool is usually the one that fits the workflow and the data constraints.

Coding: How I’m Using AI Day-to-Day (Work vs. School)

A quick bit of context: I do software development for work, but I’m also enrolled at Northeastern University’s Roux Institute in a Master of Science in Information Systems program with a focus on AI. One of the challenges is that my schoolwork has me coding in languages and frameworks that are completely new to me. For this the “AI assist” use case looks a little different than it does at work.

At work: Copilot (real-world constraints, real-world wins)

For work development, I stick to Copilot. A lot of the coding I do for work is the same as I’ve been writing for the past 25+ years, but Copilot still helps because I can give it a very specific example of what I want and save myself time.

Where it’s been especially useful for me:

• Finishing and refining TSQL - Taking something that’s 80% there and getting it over the line.
• Converting JSON into relational tables - Or generating the parsing/query pattern quickly.
• Turning a concrete example into repeatable code - Especially when the request is very “do it like this”.

I still treat the output like I would from a fast junior dev: helpful and quick, but something I verify before it goes anywhere important.

In school: Gemini + Claude as “on-demand tutoring” for new languages

For schoolwork, I’ve leaned heavily on Gemini (Enterprise) and Claude. Because I’m working in areas that are new to me, the biggest value isn’t just generating code, it’s shortening the learning curve:

• Understanding assignments and turning them into a plan – Not only do I ask AI to assist with a development plan, I give it my completed work to get a second opinion on whether I’ve met the assignment’s objectives.
• Learning new syntax/patterns quickly - With examples I can run and modify.
• Debugging support - That explains why something broke, not just what to change.
• Filling in gaps - When a lecture assumes background I’m still building.

These tools speed up my learning, but they don’t replace mastery. I still need to understand and own the work.

Claude Code (on my list to try)

One other thing that’s on my radar: I recently saw a demo of Claude Code and it looked legitimately powerful. The workflow wasn’t just “explain this code.” I’ve watched people bring in a spec document, use Claude to refine the spec, and then use Claude Code to build out a solution by pulling patterns and code samples from an existing Visual Studio project. I haven’t tried it yet, but it’s exactly the kind of “AI inside the workflow” that feels like the next step.

Reality check (because it matters)

AI can be confidently wrong. I validate outputs, test code, and cross-check assumptions. It’s a multiplier, but not a substitute for understanding.

Also: for everything AI has helped me with in Python, Java, JavaScript, DAX, and TSQL, it still can’t seem to give me a good curry recipe. No tool is flawless.

Lessons Learned

It’s been over a year of using these tools seriously—Copilot in the Microsoft ecosystem, Fabric and Power BI for data/reporting, plus other AI tools for support workflows and learning—and the pattern is clear: AI is evolving to the point where I can use it as a teammate.

The biggest takeaway is that integration matters more than hype. The best results show up when AI is embedded directly in the systems where the work is happening. That said, it still needs to be paired with guardrails and human review.