Agent Orchestration

May 2024

While much effort is being put into building one agent/ LLM to rule them all, the pursue of autonomous agents is really about laying the foundation of agent orchestration. In order for them to be more than just glorified chatbots agents need to be able to course correct, adapt to new situations and ultimately reconfigure themselves to more optimale behavior.

Not only that, they need to be able to navigate in an ecosystem that support their decisions mostly through delegation to other agents who then in return delegate to others or deliver their results back.

Finally there are many things that makes absolutely no sense to spend tokens on such as search&replace, copy&paste and query the data in a database, just to name a few. These are much better and faster utilized as functions using classical computing.

This means that agents primary function will be to figure out when it's required to use it's ability to qualitatively interpret what it's dealing with and when it should just let the classical computer do it's thing.

This is not much different to how humans use tolls for things tools are better at. In order for that to happen, language, not UI, have to be at the core of what this orchestration needs to be based on cause only language (and thus LLM) allow agents to communicate regardless of their type, destination and functionality.

Check the videos attached to see an example of how agents can be created and how they can be utilized.