AI Disruption

AI Disruption

How We Build Powerful AI Agents with Claude

Build powerful AI agents with Claude. Learn how giving it a 'computer' enables reliable task-solving in finance, research, and customer service. Discover our SDK and agent-building methodology.

Meng Li's avatar
Meng Li
Oct 09, 2025
∙ Paid
4
3
Share

“AI Disruption” Publication 7900 Subscriptions 20% Discount Offer Link.


AI Agents Are Hot, But How Do You Build One That’s Truly Reliable and Capable of Solving Complex Tasks?

Our answer: Give it a computer.

Over the past year, we’ve found that this approach has not only completely transformed programming tools but also opened up entirely new possibilities in finance, customer service, research, and other domains.

This article will share our internal methodology and best practices for building Agent loops.

Last year, we shared our experience building high-performance agents.

We use Claude for deep research, video creation, and meeting notes—it powers virtually all of our core agent loops.

In other words, the underlying capabilities that drive Claude Code can equally power any other type of agent.

In this article, we’ll dive deep into:

The core design principles behind our Agent SDK.

This post is for paid subscribers

Already a paid subscriber? Sign in
© 2025 Meng Li
Privacy ∙ Terms ∙ Collection notice
Start writingGet the app
Substack is the home for great culture