Diving a little deeper into my AI workflow, here are some more granular details.
I use several commands (or skills), to have the AI perform a series of instructions. Commands begin with a “/”.
/start {Trello URL}
Reads the Trello card from the URL and tries to understand what I want. If it doesn’t understand, it asks questions. We talk back and forth until we are both happy. Then it prompts me to approve the design.
It builds the things, and I approve the deployment to Salesforce. A command is not needed here, since it’s only 1 instruction.
Once I unit test and confirm the build, we move to the next series of commands.
/card
Updates the Trello card with everything that’s been done. Since the design and implementation may have changed along the way, this is an important step to keep things fresh.
/merge
Merges the changes into Git. The AI first creates a branch using the Trello label (feature, update, or bug) and the Trello card number. Example: “feature/NPC-123”. Then it merges the branch into the master branch.
Note: If you don’t understand branch management, just know that all changes are tracked and tagged in a source code repository so it’s easy to see what was done for each card.
/doc
Creates or updates the related client documentation. These .html docs are stored on my local computer, and I manually copy & paste them to a public knowledge base when necessary. I don’t want the AI to update them directly.
/clear
Clears the AI’s recent memory, so it’s ready to begin the next card.
The takeaway
I have a few other commands planned, but these work well for now.