Replacing entire companies with agents?

Plus: an investment fund for agent startups, an open-source LLM for visual agents, and more

Author

Michael Cunningham
January 30, 2025

In partnership with

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

Agents are great for optimizing your scripts, but make sure you optimize your agent first. If only there were an agent for that 🤔

In today’s issue…

  • A holding company replaces entire firms with AI agents

  • Run a visual agent on your browser

  • Self-building agents: teaching AI to teach itself

  • The executive guide to agent pricing

…and more

📰 NEWS

Source: adapted from google.dev

The agent orchestration powerhouse teamed up with Yohei Nakajima, creator of early agent BabyAGI to create a semi-autonomous fund to make $50-100k investments in promising AI agent startups.

Rocketable is a YC-backed holding company with the bold business model of acquiring existing companies and replacing all their human employees with AI agents.

Business process automation provider ServiceNow has been making an intense pivot towards agents, which its CEO discussed in a recent interview. The company is now launching a new orchestrator to help customers coordinate their agents, thousands of new pre-built agents, and a studio platform to create and customize them.

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

🤝 WITH ARTISAN

Automate Prospecting Local Businesses With Our AI BDR

Struggling to identify local prospects? Our AI BDR Ava taps into a database of 200M+ local Google businesses and does fully autonomous outreach—so you can focus on closing deals, not chasing leads.

Ava operates within the Artisan platform, which consolidates every tool you need for outbound:

  • 300M+ High-Quality B2B Prospects

  • Automated Lead Enrichment With 10+ Data Sources Included

  • Full Email Deliverability Management

  • Personalization Waterfall using LinkedIn, Twitter, Web Scraping & More

🛠️ USEFUL STUFF

Source: Qwen

The latest flagship model in Alibaba’s Qwen series, 2.5-VL is purpose-built to act as the backbone for computer-controlling agents similar to OpenAI’s Operator and Anthropic’s Computer Use.

mcp-agent is a simple framework intended to allow developers to easily build agents which interface via Anthropic’s Model Context Protocol, which aims to be the universal standard for agent-agent and agent-software communication.

Powered by Browser Use, WebUI enables you to launch an interactive visual agent that controls your browser, making it a free and open-source alternative to OpenAI and Anthropic’s computer use agents for web browsing.

If you prefer to build a browser agent yourself, Browserbase’s Stagehand decomposes web browsing into atomic actions—act, extract, and observe—that such agents can perform on webpage elements to complete their assigned tasks.

💡 ANALYSIS

Early agent creator Yohei Nakajima | Source: LinkedIn

Yohei Nakajima introduces a four-level hierarchy for agents that can improve their own capabilities, arguing that such agents should slowly be given more responsibility as they prove themselves, just as human employees are, in order to avoid catastrophic mistakes.

This article extends the concept of user experience (UX) and developer experience (DX) to AI agents, making the case that, when agents become the primary users of many applications, companies that build software accessible to them will have a competitive advantage over those that don’t.

An overview of the different pricing strategies being explored by agent providers and how to choose the right one for your business.

A new study of over 1,000 enterprise leaders by Salesforce found that 93% have implemented or will implement AI agents soon, but data integration has emerged as a major pain point.

Traditional request/response-driven systems, in which components act one-at-a-time, will increasingly be displaced by more flexible event-driven ones, where components can operate at any time in response to being called, this piece argues.

🧪 RESEARCH

Example of a possible computer control agent task | Source: arXiv

A comprehensive overview of the rapidly-growing field of computer-control agents (CCAs), including a taxonomy of such systems and a discussion of their challenges and researchers’ efforts to overcome them.

This paper tests the behavior of agents built on different LLMs on the iterated prisoner’s dilemma, finding that each LLM has a distinct “personality” that influences its strategy, with implications for their usage in real-world systems.

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