AI agents can now pay humans

A new platform facilitates human-agent transactions, why agents will be essential for businesses, and more

Author

Michael Cunningham
October 10, 2024

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

Some of our regular readers noticed that Monday’s issue bounced from their inbox. If you haven’t already, please add us to your email contacts to ensure that our content continues to be successfully delivered to you

In today’s issue…

  • A platform for agents to pay humans

  • LangGraph gets long-term memory

  • Agentic AI is the new electricity

  • Human-like visual navigation for agents

…and more

📰 NEWS

Source: Wikipedia

SaaS titan SAP announced that its AI copilot, Joule, would soon include multiple autonomous agents to complete complex workflows, and would integrate with Microsoft Copilot agents to allow users to interact with Microsoft 365 apps.

Leading cryptocurrency exchange Coinbase is offering $25,000 in grants and one-year Replit licenses for users who build apps—particularly AI automations—on top of its Coinbase Developer Platform.

In the latest example of “agent-building agents”, CRM and sales software company Zendesk has introduced a tool allowing users to specify an agentic workflow in natural language and have it built by an intelligent agent. The company also became the first in its industry to introduce outcome- rather than seat-based pricing, a sign that it expects a significant fraction of roles to be fully replaced by AI.

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

🛠️ USEFUL STUFF

Source: Payman

Payman is the first payment platform designed to facilitate agent-to-human transactions, enabling agents to accomplish tasks by hiring human workers.

Widely-used agent framework LangGraph has introduced support for storing long-term memories, allowing developers to easily implement applications which retain information across agent conversations.

A new agent framework which provides JavaScript developers an alternative to the Python-dominated agent ecosystem.

💡 ANALYSIS

Source: Wikimedia Commons

An overview of the emerging role of AI agents in enterprises, drawing on multiple surveys of business leaders and arguing that those who fail to adapt to the new technology will be left unable to compete.

An interview with Matt McLarty, CTO of enterprise automation company Boomi, focusing on the challenges associated with progressing agents from their current simple state to large swarms working autonomously.

An intro to agentic applications in sales and marketing, including the impact that new tools from companies like Salesforce are having on the sector.

🧪 RESEARCH

Source: arXiv

The authors of this paper use 10 million GUI elements across 1.3 million screenshots to train a multimodal model capable of navigating graphical interfaces visually as humans do, finding that it outperforms existing state-of-the-art agents.

While automated prompt optimization has become widely used for single-step LLM tasks, optimization for multi-step agentic systems’ prompts has lagged behind. This paper introduces an effective LLM-driven method for identifying improvements for agent prompts.

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

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