I’ve been testing a new way of work.
Agents are more reliable than ever. If you work with a good stack (Typescript) and have good Cursor Rules, you are coding only with prompts by now. The problem space is on the planing now.
I believe that a Trelo-like data structure is ideal to coordinate agent and human work. Github Projects is the best place for that because it’s adjacent. I’ve been using Github Issues to describe, in text, the feature we want and that will serve as prompt eventually.
The secret is iteration over a plan.
That’s what I need from BugBot:
I want to customize BugBot to act as a code planing copilot.
Because it runs on my repository, it has all the rules and knowledge about the current implementation. All I need is custom prompts + UX thinking to make it work.
The workflow would be something like:
Open Issue (dev, customer, enduser) →
BugBot automatically thinks on a plan and comment →
Dev/stakeholders comment on the plan →
BugBot updates the document →
Dev approves execution →
BugBot creates the PR and comments Deploy Preview URL →
Dev/stakeholder comment on the UX →
Feature shipped