Overcoming Microsoft Copilot Studio Limitations: Building an AI Image Comparison Agent

While working on an AI-powered comparison agent, I came across a limitation that pushed me to rethink the overall architecture of the solution.

The idea behind the agent was simple — users should be able to upload two versions of a product label, and the system would automatically compare images using AI and highlight the differences.

The most useful feature I wanted to build was AI image annotation. Instead of just showing text results, the system should visually highlight the exact areas where changes occurred on the label.

For example, if an ingredient or nutrition value changed, the system should clearly mark that section so users could quickly identify what was modified. This is where an AI image comparison tool becomes more practical than plain text output.

The Challenge

I initially built the agent using Microsoft Copilot Studio, as it is a strong platform for conversational experiences and common Copilot Studio use cases.

It worked well for:

· Handling user prompts

· Managing file uploads

· Returning text-based responses

However, I soon understood the limitations of Copilot Studio.

It could not:

· Perform image processing

· Handle automated image comparison at a visual level

· Generate annotated images with highlighted differences

In short, while it could explain differences, it could not act as a complete AI image comparison solution.

Rethinking the Solution

To overcome this, I redesigned the architecture by combining multiple services instead of relying on a single tool.

Using Azure Functions for Image Processing

I created an Azure Function App to handle the core logic. This is where Azure Function image processing comes into play.

The function:

· Compares two images

· Detects differences

· Adds visual highlights directly on the label

This made it possible to build a proper automated image comparison workflow.

Using Power Automate for Integration

Next, I used Power Automate to connect everything. This is a practical example of Power Automate integration and workflow automation tools working together.

The flow:

· Takes files from Copilot Studio

· Sends them to Azure Functions

· Receives the processed image

· Returns the final annotated result

Copilot Studio as the Interface

Finally, Microsoft Copilot Studio continues to handle user interaction.

It:

· Accepts user inputs

· Displays results

· Keeps the experience conversational

Final Architecture

· Copilot Studio → User interaction

· Power Automate workflow → Integration and orchestration

· Azure Functions → Image processing and annotation

Key Takeaway

This experience clearly showed that building strong AI solutions often requires combining tools.

Instead of forcing everything into one platform, it is better to design the right architecture using the right services. By integrating Copilot Studio with Azure Functions and Power Automate, it is possible to overcome platform limitations and build a complete AI image comparison and annotation solution.