The AI Codes, the Human Just Decides and Deploys: Inside Kune's Workflow

Sergey Kune has published the workflow he built to cut down on the fatigue he felt while using AI coding tools. The core idea is simple: pull out only the things a human actually has to decide, and turn them into a prioritized list. Kune says he uses the system himself and keeps refining it as he goes.

Every task gets its own worksheet, stored as a single file inside the code repository. The worksheets are written in Markdown (a format that organizes documents using simple symbols), and the overall status of all work is tracked in Git, the tool that manages change history. The agent, an AI that handles work on its own, loops through research and verification. For implementation, it only touches tasks a human has marked as safe. Whenever something comes up that requires human judgment, that item gets stacked onto the list. Kune says he reviews them one at a time and records each decision back in the worksheet.

Sensitive code changes go through a separate approval step. According to Kune, those changes are made on a branch with its own isolated workspace, and then submitted for approval. Deployment, the step that pushes changes to the live service, is something Kune handles himself.

Kune says there are still rough edges. Tasks waiting on a decision get blocked over and over, and it is hard to grasp the context written in a worksheet quickly. He says he is working on fixing that. In the end, collecting the items that need review is not where it stops. What remains is the challenge of helping a human understand each item quickly and actually reach a decision.