Build a Multi-Agent Solution within the Claude App using this INSANE feature

Claude has recently introduced a powerful new feature to its app that enables developers and creators to use AI to build robust applications with just a few prompts, without ever having to leave the Claude app.

This makes it especially easy for citizen developers to build and share AI-powered apps without the risk of paying for API calls (you will read more about this later).

You can find the official announcement here: https://claude.com/blog/claude-powered-artifacts

In this blog post, we will use this feature to take things a step further and build a simple multi-agent (or, to be more precise, multi-LLM) solution with a single prompt.

Let’s dive in.

How it works

At its core, this feature works through an API call to Claude that is executed on behalf of the app's user, not the developer. This allows for the creation of impressive applications.

But can this same feature be used to make multiple, coordinated LLM calls? Can you create a "multi-agent" system where different AIs work together to break down a large task or provide specialized input to solve a complex problem?

The answer is: ”yes."

While parallel execution is a field for further exploration, this capability already opens up a world of possibilities.

Example: Advisory Board

Imagine you have a business idea and want to gather opinions from various experts, e.g. the tech and the business experts. All with a single prompt.

The resulting app could look something like this:

To keep things simple, we'll create an app that aggregates summaries. It will make two separate API calls to Claude to generate two distinct summaries of a topic and then synthesize them into a final summary (with a third call).

Step by step: Build a multi-LLM app with Claude

First, provide a description of the purpose and content of the app you want to create. Then, add the prompt provided at the bottom of the post.

I had to run it multiple times to execute multiple calls, and the prompt below will allow Claude to use the approach that worked for me. Feel free to try other ways if you wish.

Here is the complete prompt to feed Claude:

In just a couple of minutes, your new application will be ready to use and publish.

I tested it with a random topic, and as expected, the app generated two completely separate summaries:

Followed by the final, synthesized summary. The result is quite impressive.

What Else Can You Build With This Technique?

The potential applications for this multi-agent approach are vast.

Here are just a few ideas generated by Claude.

*   **Virtual Advisory Board:** Simulate a board of directors with different AI personas (e.g., CEO, CFO, CTO, and a marketing expert) to get diverse feedback on business strategies.
*   **Complex Task Breakdown:** For a large project, one AI agent can act as a project manager, breaking down the main goal into smaller tasks, while other specialized agents can then tackle each sub-task, such as coding, writing documentation, or creating marketing copy.
*   **Content Creation Team:** Assemble a team of AI agents where one is a researcher, another is a writer, a third is an editor, and a fourth is a social media specialist, to generate comprehensive and ready-to-publish content.
*   **Personalized Education Tools:** Create a learning experience where one AI acts as a tutor explaining a concept, another generates practice problems, and a third provides feedback and encouragement based on the user's performance.
*   **Financial Planning System:** Design a system where one agent analyzes a user's financial goals, another researches investment options, and a third provides a tailored financial plan, all while a fourth agent monitors the market for relevant changes.
*   **Game Development Assistant:** Have one AI generate the game concept and story, another write the code for the mechanics, and a third design the user interface and assets.

If you're feeling inspired to create something cool, please share it in the comments. Thanks! :)