The article discusses Embabel's innovative framework that integrates a planning step to derive actions from application code using a deterministic AI algorithm. This approach focuses on creating a rich domain model, largely through Kotlin data classes or Java records, ensuring that prompts are type-safe and maintainable. Johnson highlights the necessity of a higher-level orchestration technology beyond the Model Context Protocol, citing requirements for safety, explainability, and integration with existing systems. His ambition is to not only establish Embabel as a top agent platform on the JVM but as the best overall.
Embabel introduces a planning step that discovers actions and goals from application code, using a non-LLM AI algorithm for deterministic planning.
Building a rich domain model in applications ensures that prompts are type-safe and can be refactored, allowing for additional behavior in domain objects.
Johnson emphasized the need for a higher-level orchestration technology, noting aspects like explainability and composability of flows as essential for safety and integration.
"We want not just to build the best agent platform on the JVM, but to build the best agent platform, period," Johnson stated.
Collection
[
|
...
]