Agentic AI for product operations
A Global 500 consumer products company piloted four agents in Azure DevOps. Product leaders in the pilot spent less time maintaining the backlog.
Sigao built the pilot with Microsoft Foundry and its agentic framework for product management.
The backlog had become too messy to plan from.
Nobody could see where effort went
Work, requests and technology investments sat in separate systems.
The backlog outgrew its upkeep
Thousands of items had piled up, many of them duplicate, stale or orphaned.
Requests came in from everywhere
Each channel delivered a different level of quality and context.
Planning turned reactive, and product leaders spent too much of their time on administration.
We began with the people who owned the problem.
Form an AI community of practice
Early adopters among the product leaders met for instruction, coaching and feedback.
Find where time was being lost
The group named the tasks that cost product leaders the most time.
Build agents and the process around them
Each agent came with an operating process, aimed at the most painful work first.
Four agents sit between a request and a prioritized backlog.
Intake agent
Turns emails, meeting minutes and market research into structured backlog items.
Routing agent
Sends each item to the team with the right skills, product knowledge and availability.
Estimation agent
Sizes the work from the team's own backlog history and reference stories.
Prioritization agent
Weighs estimates against business value and recommends a new backlog order.
The agents change the backlog directly, under the client's rules.
For example, the intake agent turns a request from meeting notes into a structured Azure DevOps work item. The routing agent uses the teams’ skills, product knowledge and availability to direct it. The client’s permissions govern which backlog changes the agents can make.
In the pilot, backlog upkeep became automatic and continuous.
Product leaders in the pilot also reported more confidence in planning and priorities.
The client's name is withheld at its request.
| Before | After |
|---|---|
| Manual backlog intake | Automated intake from several sources |
| A reactive backlog | Continuous backlog hygiene |
| Limited visibility | Context-aware insight |
| Time serving the tool | Time serving the product |
Tell us where delivery is getting stuck.
If your backlog looks the way this one did, start with a conversation about how work moves through your teams today.