integrating agent chat/search/suggest into development workflow
Agents and LLMs can speed up a users output considerably
over manual-only effort. Since every developer has preferences and different strengths,
one needs to try a few different approaches to optimize for an AI assisted workflow.
Season, Generate, Refine Approach
One approach for effective AI support workflows is separate phases of development that better
suit manual vs automated code generation. This approach yields better success because the structure and goals are
defined simply initially with scope constraints and only adding additional complexity for the AI at subsequent steps.
Finally a manual review/refinement is executed to tighten up areas and polish the execution.
⟡ Season
Create a simple UI using static test data.
⟡ Generate
Create a data structure matching the working model.
Build a backend to facilitate relational and CRUD operations.
⟡ Refine
Improve UI/UX design and interactions.
Add validation to user input.
Add basic code coverage.
This method is by no means plug-and-play and will require re-prompting to yield
the desired results. Additionally crafting the best prompts will require knowledge
of the frameworks and best practices keep the agent solution true to your request.
This will however allow you to create simple functional starting points at least
twice as fast if not more. This is a skill that will come to be expected just as
using search engines or reference books has been in the past.