applications

Mixture of Agents

A system architecture where multiple LLMs collaborate by generating and refining responses iteratively. Each agent may have different strengths, and their outputs are aggregated or refined to produce a higher-quality final response.

In practice

A mixture of agents might have one model draft a response, another critique it, and a third refine the final output.

Multi-Agent System
A system where multiple AI agents work together, each with specialized roles or capabilities, coordinating to solve complex problems. Multi-agent systems can tackle tasks too complex for a single agent by dividing work and sharing information.
Mixture of Experts
An architecture where multiple specialized sub-networks (experts) exist within a model, with a routing mechanism that activates only a subset of experts for each input. This allows models to have enormous total parameter counts while only using a fraction of compute per inference.
Agentic AI
AI systems designed to act with agency, taking initiative to accomplish goals through planning, tool use, and iterative reasoning. Agentic AI represents a shift from passive response generation to proactive task completion.