Multi-Agent System
A multi-agent system is a software system composed of multiple cooperating AI agents, each with a role, tools, and memory, coordinating to solve a shared goal.
In plain English
A multi-agent system is what happens when you stop trying to cram every capability into one giant prompt and instead split work across specialists. One agent plans. Another writes code. Another reads web pages. Another checks quality. Each has its own system prompt, its own tool belt, and often its own short-term memory. A coordinator — sometimes called a CEO, manager, or router — decides who does what and in what order.
The win is focus. A specialist with a short, targeted prompt and three tools will outperform a generalist with a sprawling prompt and thirty tools on the same task. The cost is coordination: you need a way to pass results between agents, handle failures, and avoid infinite loops. Modern frameworks (Claude Agent SDK, LangGraph, AutoGen, CrewAI) exist precisely to solve that coordination problem.
Why it matters for Black Box
Black Box is a multi-agent system by design. A CEO agent receives the goal and delegates to 18 specialists — covering coding, content, research, browser, business ops, design, product, marketing, sales, social, SEO, ads, PR, customer success, support, data, finance, and legal — each with a role-shaped prompt and the tools to match. The owner never has to orchestrate anything; the system does.
Examples
- A CEO agent asks a Research specialist to gather sources, then hands the output to a Content specialist to draft a post.
- An Evaluator agent rejects a draft and routes it back to the Content agent with specific feedback.
- A Browser agent logs into a vendor portal and passes the scraped data to a Business Ops agent for reconciliation.