My first foray into the UI Expert Circle…I’m wondering if there’s a place for leveraging code agents (like Claude, for example) to develop and deploy custom DX components in Constellation. For example: being able to provide a business description of the control/capability needed and having Claude reason through available components for a match, for extensibility, or if a net-new component is needed based on Constellation’s design system, and then executing the necessary steps to finalize the component for deployment into the environment.
I can see this being a huge step toward Constellation adoption for customers. One of the biggest gaps I still see with customers is fear of adopting Constellation because of the standardization. Giving up control and flexibility is a big concern for technology teams, so providing a way to give them extensibility that doesn’t sacrifice the design-system standards would be huge, especially if it could be provided in a “Blueprint-like” interface that would allow some visualization of the component’s behavior within the context of a case.
Thomas, just posted an article related to your ask about using code agents to create DX Components. The possibility is there while acknowledging @RameshSangili points about
creating tech debt in the future.
That’s always a concern with “customization” vs. configuration for sure…but I think there’s a fine line where there’s some critical needs that Constellation doesn’t address today where having the ability to manage the DX component process through code-assistants might really help accelerate builds where one critical UI gap might derail a project. But like you said, I see this as being a process that needs some key governance so that it doesn’t go wild.
Agree with you and @RameshSangili - AI certainly makes this more accessible and a viable path but a balance needs to be struck.
The other path is going SDK?
Maintaining a working OOTB Constellation application with your own SDK version? The problem with overloading with custom components is if one breaks then its critical path for a release, you’ve mixed OOTB platform with your own code and everything must work.
With an SDK implementation, there is a separation of concern and always a working app, with SDK implementation a follower. This could be a way to strike that balance? Even before generative AI hit the level it is today, I know of some clients that had both.
I am not sure if that is better or easier to implement with the newer tools today but certainly an option to consider in this brave new world.
