Two monitors at night: a website design on the left, its sections linked by glowing lines to task cards on the right, an approval stamp on the desk

From Figma to a full Azure DevOps backlog - with a human approving every page

How we used an AI agent connected to Figma and Azure DevOps to turn a design into a complete, reviewed backlog - and then into working pages.

Development 5 min
Mark Podliashetskyi
Mark Podliashetskyi CEO
Share:

For our new company website, an AI agent read the Figma design and turned it into a full Azure DevOps backlog - epics, user stories and tasks - and then built the public pages from that backlog. The agent did the decomposition; a human approved every page before anything reached the backlog.

What problem does this solve?

Turning a design into a backlog is slow, careful work. Someone has to go through every page, every block and every state, and write tasks a developer can pick up without asking questions. It is necessary work, and it is exactly the kind of structured, repetitive work an AI agent does well - as long as someone who knows the product checks it.

How did the setup work?

We connected two systems to the coding agent (Claude Code) through MCP - the Model Context Protocol, a standard way to give an AI agent access to external tools:

  • Figma, so the agent could read the actual design: pages, sections and components.
  • Azure DevOps, so the agent could create work items in our real backlog.

The workflow, step by step

  • The agent read the Figma design module by module and page by page.
  • For each page it wrote tasks with descriptions and links back to the design - as local files first, not in Azure DevOps.
  • I reviewed each page's tasks and approved them, or sent them back with corrections.
  • Only approved pages went further: the agent created the epics, user stories and tasks in Azure DevOps.
  • The agent then built the public-facing pages of the website from those same tasks.

Why approve page by page instead of at the end?

Because a backlog is much easier to fix before it exists. The agent wrote to local files first, so a misread design cost me a comment, not a cleanup of dozens of work items. Approving one page at a time also kept each review small enough to do properly. Reviewing an entire generated backlog in one sitting is how you end up approving things you did not really read.

It also meant the backlog was one I could stand behind. Every item in Azure DevOps had been seen and accepted by a person who knew what the page was supposed to do.

The agent did the decomposition. The decisions stayed with people.

- Mark Podliashetskyi, CEO

What did we get at the end?

A complete, structured backlog for the website - epics, user stories and tasks with descriptions and links to the design - and then the website's public pages, built from that backlog. Because the tasks pointed back to Figma, anyone picking one up could see the exact design it came from.

Where does this fit for your project?

This works best when the design is reasonably final and the team already uses Azure DevOps or a similar tracker. It does not remove the project manager - it changes their job from typing tasks to judging them. That is the pattern behind how RemSoft.Dev delivers with AI: agents do the volume, and a human approves every step that matters.

Want to Talk to Our Team?

Our engineers can help map this to your stack.

An unhandled error has occurred. Reload 🗙