From feature request to release, with human approval

The challenge

Every feature passed through at least three handovers, with queues between them. There was no single place to coordinate those transitions, so context and decisions got lost as work moved from one person to the next.

AI hadn’t changed that dependency. Progress still relied on a specific person in each role being available to prompt the AI, review its response and ask it to continue. We needed a clear path for each feature from request to release, with agents moving the work forward and humans reviewing, adjusting and approving the results along the way.

What we built

An agent that guides each feature through 12 steps, including intake, problem research, solution options, prototype, breakdown, plan, build, code review, QA, release and post-release follow-up. Every step ends with an Approve / Decline / Refine gate for a human.

• Specialised subagents: code reviewer, “ripple” reviewer (side effects on other parts of the system), legal, market and data researchers, prototype reviewers.
• Integrations through MCP: Jira and Confluence, Airtable, Playwright, the company’s own business-operations MCP server, plus read-only database access and on-demand sandbox environments.
• Where people already work: stakeholders send requests from Microsoft Teams. A bridge runs the intake agent headless and posts the triage outcome back to the thread.
• Memory you can audit: each feature has a folder with a dated decision log, and the agent is the single writer to Jira.

Results

The system brings the delivery process into one coordinated workflow, with four key benefits:

• Less waiting between steps. Once a person approves a stage, the agent can take the next step without waiting for someone to start another AI session.
• Less context lost at handovers. A shared feature record and decision log should reduce repeated explanations, missing requirements and work redone because earlier decisions were lost.
• Less dependence on one person in each role. A defined workflow should make it easier for another team member to pick up a review with the context they need.
• More human attention on decisions. People can focus on reviewing evidence, adjusting the direction and approving results, with less time spent coordinating handovers and prompting each step.