I tested langchain4j agentic platform and the main difference I see is configurability.

The langchain4j agentic platform is fully declarative, meaning that you have to declare your agents in interfaces in pure java code with annotation syntaxes, and this seems to be really enforced, to the point that you can't define purely your agent with builders.(for example, defining a "User template" was enforced with annotations for example).

I haven't seen this with ADK, where you can fully configure your agent with their builders and pick the way you instantiate your agents.

This is a crucial point if you need to configure your agent from a database or configuration files for example.

This clearly delineate a separation of concern between a fully "annotation" based framework, vs a core framework that works on builders, that seems more clever, because it doesn't prevent to build the declarative implementation based on that core framework later on.

Declarative frameworks looks more like playground than really matured frameworks - even though you could expect to have your agent code versioned in your repository, with traceability, etc, it doesn't work when your application offers agents configurability, with quick iterations etc...