Anthropic let an agent run a real business

Plus: what agent buyers want, an AI-powered coding platform for beginners, and more

Author

Michael Cunningham
July 03, 2025

Welcome back to Building AI Agents, your biweekly guide to everything new in the field of agentic AI!

I though the path to AGI was leaving OpenAI and raising billions in VC funding to supposedly build superintelligence but never actually releasing anything.

In today’s issue…

  • The skills and automations agent buyers are looking for

  • An AI-powered coding platform for beginners

  • Anthropic let an agent run a real business

  • Coding agents 101

  • The future of software is “agents all the way down”

…and more

Building AI Agents’ Official Course and Community

Learn how to…

  • Build and deploy agents using tools that require zero coding experience

  • Identify pain points in your business or personal work ripe for AI

  • Sell agentic automations to businesses

…while networking with other agent builders and entrepreneurs

“Agents are not only going to change how everyone interacts with computers.

They’re also going to upend the software industry, bringing about the biggest revolution in computing since we went from typing commands to tapping on icons.”

—Bill Gates

📰 NEWS

Source: LlamaIndex

LlamaIndex, already one of the most popular agent frameworks, rolled out Workflows 1.0, a lightweight package for building agents in Python and TypeScript.

Cloudflare, which hosts around 20% of the internet, released Pay per Crawl, a marketplace in which websites can charge AIs fees for reading their content. The project addresses site owners’ concerns about AI scraping—and creates a potentially lucrative business model for them to charge AI agents for access.

If you find Building AI Agents valuable, forward this email to a friend or colleague!

🛠️ USEFUL STUFF

Source: Pintrest

The author of this roundup scraped over 1,000 Upwork posts seeking freelance agent builders, identifying the most in-demand skills and most common pain points.

Google’s Colab platform, which provides free computing for those experimenting with programming, now has built-in agentic capabilities that allow users to generate and modify their code with AI.

Michael’s note: given how powerful—though by no means essential—coding is for agent-building, Colab is a great resource for getting your feet wet with Python programming. I built many of my first AI projects in Colab.

A guide by Cognition, creator of the viral Devin coding agent that arguably launched the agents-as-software-engineers trend, to building agents capable of advanced programming.

💡 ANALYSIS & RESEARCH

Source: Anthropic

The AI company ran an experiment in which it allowed an agent powered by its Claude LLM to manage a real vending machine in its office, setting prices and sending emails to have the machine restocked. Although the agent went out of business, Anthropic pointed out that there are clear paths to addressing many of its shortfalls, and plans to try again.

Headless browsers—browsers which lack a graphical interface—are a critical technology that enables agents to browse websites, even when they don’t expose an API.

“Context engineering” is the art and science of providing LLMs with all the necessary information to perform their task, while avoid any un-necessary detail. This article breaks down some of the common methods for giving agents the ability to “remember” critical context.

Bonus: this blog post by LangChain founder Harrison Chase, also on context engineering, and how the company’s tech enables it.

Riding high on the success of his company’s Replit agent, which creates (usually) fully-functional apps from a simple request, CEO Amjad Masad envisions a future where everyone can easily build software, regardless of technical skill.

This paper breaks down the three major types of agent communication protocols, and compares the most popular ones.

Thanks for reading! Until next time, keep learning and building!

What did you think of today's issue?

Login or Subscribe to participate in polls.

If you have any specific feedback, just reply to this email—we’d love to hear from you

Follow us on X (Twitter), LinkedIn, and Instagram