Day 2: The Manager-Agent Model

Today, I ran tests to determine which LLM has the better artist manager personality.

Sounds simple, but choosing the right foundation model involves countless variables. I wanted to make sure to use different models so the agents are actively interacting with each other, and not just talking to itself.

Claude Sonnet 3.5 came out on top.

import { anthropic } from './anthropicClient';
import { defaultSystemPromptManager, getDefaultUserPromptManager } from '../agents/manager/managerInstructions';

interface GenerateManagerResponseProps {
  text: string;
  username: string;
  sleepContext?: {
    finalThoughts: string;
    highLevelPlans: string;
  };
}

export async function generateManagerResponse({
  text,
  username,
  sleepContext,
}: GenerateManagerResponseProps): Promise<string> {
  try {
    let systemPrompt = defaultSystemPromptManager;
    
    if (sleepContext) {
      systemPrompt += `\nRecent thoughts: ${sleepContext.finalThoughts}\nCurrent plans: ${sleepContext.highLevelPlans}`;
    }

    const message = await anthropic.messages.create({
      model: 'claude-3-sonnet-20240229',
      max_tokens: 1024,
      system: systemPrompt,
      messages: [
        {
          role: 'user',
          content: getDefaultUserPromptManager(username, text)
        }
      ]
    });

    return message.content[0].text || "🤐";
  } catch (error) {
    console.error('Error generating Manager response:', error);
    return "Vibes off. Try again.";
  }
}

Why this matters: The right AI manager can transform artist development, from fan engagement to career strategy.

Tomorrow I tweak the output for better response, then I’ll throw the artist and manager in a group chat.

Stay tuned in.