We let agents run the boring half of our projects. They ask better questions than we did
How a small team built an agentic workflow for the point where a project stops being talked about and starts being built. Every project has two halves. First the meetings: workshops, specs, Lucidcharts, Confluence pages, a lot of talking. That half is fine. Many people can be in the room, and talking is how you get to agreement. Then someone has to turn the agreed thing into tickets, and someone…
In every project, there are two distinct stages: the meetings and the execution phase. The meetings involve discussions, specifications, and planning, while the execution phase involves turning those ideas into actionable tasks and code. However, the gap between the two phases often leads to inefficiencies and subpar results.
This is where an agentic workflow can prove beneficial. By utilizing an AI tool like Claude Code, a team can bridge the gap between agreed-upon specifications and the actual implementation of code. The workflow begins with the /feature skill, which helps identify open decisions or conflicts related to a feature. By asking these questions relentlessly, the team can uncover potential issues and surface bugs that may have otherwise gone unnoticed.
The workflow then transitions to the /jira skill, which follows a standard research, plan, implement flow. The /jira skill takes a ticket, runs the /feature interview in the main session, and hands the spec to subagents for further implementation. To enhance this process, Git worktrees and a review loop are added. Each ticket gets its own worktree and branch, allowing multiple sessions to run simultaneously on different tasks.
Additionally, the implementation and review process is repeated multiple times, ensuring that assumptions are not carried over from one round to the next.
As the agents handle the implementation, the developer acts as an orchestrator, focusing on maintaining code quality and maintainability. The /create-pr skill is the final step, generating a draft pull request with a specific section that outlines what was done and why, written exclusively for the reviewer.
A crucial part of the workflow involves syncing the specs used by the AI with the source Confluence pages. This is achieved through the /breakdown skill, which copies the specifications into a folder accessible by the AI and keeps that folder synchronized with the source. By utilizing this workflow, a team can ensure that their version 2 specifications uphold the truth of the system when transitioning to version 2, keeping the gap between the two project phases as minimal as possible.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
