One Agent To Rule Them All?

According to IBM an AI agent is: “An artificial intelligence (AI) agent refers to a system or program that is capable of autonomously performing tasks on behalf of a user or another system by designing its workflow and utilizing available tools.”

Like me, you probably start your initial AI agent architecture by defining a single agent and giving it all tools it needs, combined with a detailed prompt. There are multiple problems with this approach.

• When your agent grows in complexity, the model’s ability to consistently and accurately choose the right tool for each task becomes more challenging. 
• Additionally, certain tasks may benefit from specialized prompts or models. For instance, if the agent is addressing an upset customer, you might want it to adopt a more professional tone, while the agent greeting the customer could have a friendlier, more casual tone.

Possible solutions

One approach you can use to deal with the issues that arise when building complex agents is to create many specialized agents and use a general purpose agent to delegate tasks to them. There are specialized models (https://github.com/Not-Diamond/awesome-ai-model-routing) you can use for this delegating agent. Another approach that can simplify your setup is by leveraging an AI framework.

What are AI frameworks?

AI frameworks streamline the development and deployment of complex algorithms by offering pre-built functions and libraries. This enables developers to customize AI models for specific needs without having to create the underlying systems from the ground up. Additionally, these frameworks standardize the development process, providing teams with consistent tools and methods across different projects. A new framework created by Google is Firebase Genkit.

Firebase Genkit

Firebase Genkit (https://firebase.google.com/docs/genkit) is an open-source framework that provides a structured environment for developers to leverage the capabilities of generative AI. Genkit supports both Google’s native models and third-party open models, and is designed to be highly extensible, allowing developers to tailor it to their needs. It also integrates seamlessly with vector databases like Pinecone and PostgreSQL’s pgvector or Google Cloud’s firestore.

Genkit also offers primitives that make it a lot easier to define a multi-agent architecture. Below, a snippet from the genkit Documentation that illustrates this approach.

// Define a prompt that represents a specialist agent
const reservationAgent = ai.definePrompt(
{
name: ‘reservationAgent’,
description: ‘Reservation Agent can help manage guest reservations’,
tools: [reservationTool, reservationCancelationTool, reservationListTool],
},
‘{{role “system”}} Help guests make and manage reservations’
);

// Or load agents from .prompt files
const menuInfoAgent = ai.prompt(‘menuInfoAgent’);
const complaintAgent = ai.prompt(‘complaintAgent’);

// The triage agent is the agent that users interact with initially
const triageAgent = ai.definePrompt(
{
name: ‘triageAgent’,
description: ‘Triage Agent’,
tools: [reservationAgent, menuInfoAgent, complaintAgent],
},
`{{role “system”}} You are an AI customer service agent for Pavel’s Cafe.
Greet the user and ask them how you can help. If appropriate, transfer to an
agent that can better handle the request. If you cannot help the customer with
the available tools, politely explain so.`
);

// Start a chat session, initially with the triage agent
const chat = ai.chat(triageAgent);

I think you refer to a ‘Agentic Workflow’, as coined by the Godfather of Machine Learning.