🚀 Building a Feature Spec Assistant with Microsoft Copilot for Product Management

April 30, 2025

In early 2025, I had the opportunity to participate in a Coretek beta and pilot engagement focused on exploring practical, high‑impact applications of Copilot-style agents inside real business workflows. What started as a series of exploratory workshops quickly became a proving ground for something I had been thinking about for a long time: how to embed product strategy and discipline directly into the act of writing product artifacts.

Between March and April 2025, I attended multiple Coretek-led working sessions, including an “art of the possible” workshop and several structured use‑case mapping meetings, followed by explicit pilot check‑ins. These conversations created the space to experiment with an internal agent I ultimately incorporated into my day‑to‑day product management workflows: the Feature Spec Assistant Copilot.

🧩 The problem I was trying to solve

As a product manager, a significant portion of my impact is expressed through written artifacts:

• Product Requirement Documents (PRDs)
• Epics and feature specifications
• User stories that engineers, QA, and leadership all interpret differently

Even with strong templates and guidelines, consistency is hard:

• Business goals get diluted into technical descriptions
• Epics drift toward activity-based naming instead of outcomes
• User stories vary in quality depending on time, context, and cognitive load

I wasn’t looking for an agent to think for me. I wanted one that would reinforce how we think.

🤖 What the Feature Spec Assistant Copilot is (at a high level)

The Feature Spec Assistant Copilot is designed to act as a product management assistant inside ImageTrend, not a generic writing tool.

At a conceptual level, the agent:

• Operates as a PM-side copilot, not an engineering or support agent
• Anchors all output in existing product strategy, initiatives, and team guidelines
• Encourages outcome‑oriented thinking over task‑oriented documentation

Rather than inventing structure, it enforces it.

🛠️ How it supports core product artifacts

Knowledge Sources**

To enable this level of support, I’ve connected Copilot to a comprehensive set of knowledge sources:

• My entire product wiki in Azure DevOps
• The Teams/SharePoint folder for my product
• Our PRD templates and product team guidelines
• The PLDC (Product Lifecycle & Development Cycle) document
• Historic product strategy documents highlighting primary features
• Azure DevOps work items, accessed via retrieval augmented generation (RAG)

This breadth of context ensures the agent’s responses are grounded in real, up-to-date product knowledge and operational discipline.

Instructions

📄 PRDs (Product Requirement Documents)

When asked to help with a PRD, the agent:

• Uses an established PRD template
• Structures content around purpose, value, and measurable outcomes
• Keeps the focus on why the work matters, not just what is being built

The goal is not speed—it’s clarity and repeatability.

🎯 Epics

One of the strongest patterns reinforced by the agent is outcome‑based epic naming.

Instead of:

“Certification Management Update”

The agent pushes toward formats such as:

• “Improve ___ for ___ so that ___.”
• “Increase/Decrease ___ by ___ through ___.”
• “Deliver ___ that enables ___ to ___.”

It also prompts the same leadership‑level questions every time:

• What customer problem is being solved?
• What impact does this have on revenue, retention, or strategic deals?
• How is the business or customer measurably better when this is complete?

This consistently reframes epics as strategic investments, not delivery buckets.

🛠️ Features

For feature specifications, the agent reinforces a minimum standard:

• Clear details
• Explicit purpose tied to business goals
• A defined value measurement
• An investment category

This ensures features are discussed in the same language leadership uses to make prioritization decisions.

📝 User Stories

When user stories are needed, the agent enforces a consistent structure:

• Clear archetypes
• Action‑oriented titles
• A focus on value and justification

It also encourages disciplined descriptions that include:

• Purpose
• Permissions and setup
• Expected behavior written using a GIVEN / WHEN / THEN approach
• Fringe cases and scoped limitations

The result is fewer ambiguous stories and more testable outcomes.

💡 Why this mattered during the Coretek pilot

The Coretek beta and pilot sessions weren’t about building a flashy demo. They were about pressure‑testing whether Copilot agents could reinforce real operational discipline.

By grounding the Feature Spec Assistant in existing product guidelines and strategy—and using it during live work rather than hypothetical exercises—I was able to validate something important:

Copilot agents are most valuable when they encode how an organization thinks, not just what it knows.

🎁 Bonus: Fortuitous Copilot Moments

Going all-in on Copilot has led to some unexpectedly valuable moments in my workflow:

1. Instant Strategic Context

While working on a strategy document, I prompted the agent for supporting materials. It automatically surfaced a fresh PowerPoint from leadership, highlighting our Ideal Customer Profile (ICP) and strategic market segments—exactly the context I needed, right when I needed it.

2. Persona Documentation, Perfected

When documenting user personas, I wanted to ensure accuracy beyond my own recollections from meetings. I simply asked the agent to search through my emails and meeting notes for relevant details about people, positions, and needs. The result was spot-on—so much so that it enabled me to succinctly convey what I already knew to a broader team, making the product ask clear and actionable.

The real win isn’t just that Copilot can generate this information, but that its ability to do so empowers me to communicate with clarity and confidence across the organization.

Categories: projects

Tags: product-management, technology, systems