- Building AI Agents
- Posts
- Microsoft's new general agent strikes back at Salesforce
Microsoft's new general agent strikes back at Salesforce
PLUS: AI agents for software development and simulated AI civilizations
Michael Cunningham
November 07, 2024
In partnership with
Welcome back to Building AI Agents, your biweekly guide to everything new in the AI agent field!
If one AI agent fires another for bad behavior, does that make it a terminator terminator?
In today’s issue…
Magentic-One: Microsoft’s new open-source general agent
Amazon’s framework for running agents in the cloud
AI agents for DevOps—and what it means for human coders
Simulated civilizations with >1,000 AI citizens
…and more
📰 NEWS
Source: Microsoft
Microsoft released Magentic-One, a multi-agent system for solving general tasks, which can surf the web, write and execute code, and operate through a computer terminal. It is believed to be intended in part to address rival offerings by Salesforce and the scathing derision Salesforce CEO Marc Benioff has aimed at Microsoft’s agentic products.
Anthropic has unexpectedly raised the price of its Claude 3.5 Haiku LLM, citing its high performance, representing a rare reversal which could potentially signal the end of the era of rapidly decreasing LLM prices.
If you find Building AI Agents valuable, forward this email to a friend or colleague!
🤝 WITH ARTISAN
Hire an AI BDR to Automate Your LinkedIn Outreach
Sales reps are wasting time on manual LinkedIn outreach. Our AI BDR Ava fully automates personalized LinkedIn outreach using your team’s profiles—getting you leads on autopilot.
She 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: Amazon Web Services
An agent orchestrator by AWS which allows users to build complex, multi-agent systems by intelligently routing queries to different agents locally or in the cloud.
Skyfire is a payment network backed by a16z and Coinbase Ventures, intended to facilitate a growing segment of the AI agent ecosystem: agents which can transact with humans and each other.
A set of quick and easy tutorials on implementing four simple but powerful agent system architectures—reflection, tool use, planning, and multi-agent—without the need for 3rd party frameworks.
💡 ANALYSIS
Source: Wikipedia
This article explores the ways in which AI agents could revolutionize software DevOps—and where that leaves human engineers.
A recap of an interview with OpenAI executives Olivier Godement and Romain Huet in which Godement lays out the challenges of making AI fully agentic but argues that, in a few years, “every human on Earth, every business, has an agent”.
This piece examines why a recent study found coding agent performance on the dominant SWE-Bench assessment may be overestimated by as much as 20-fold, and argues in detail for more careful evals.
Consumers’ trust in companies is at a record low, but, according to a recent survey by Salesforce, they are open to working with AI sales agents to help address that distrust.
🧪 RESEARCH
Source: ArXiv
This ambitious effort used as many as 1,000 agents interacting with each other in a Minecraft world to simulate human civilization, successfully recapitulating societal patterns such as specialization of labor and the spread of religion. Although such work seems frivolous, it has important implications for training agents to function in real human society.
The authors of this paper solve what they term the Agent Communication Trilemma with a custom protocol for LLM agents to communicate with each other, cutting costs five-fold relative to simple natural language.
Prediction markets require an unbiased “oracle” to adjudicate the outcomes of forecasted events with perfect reliability. This piece gives a case study in how Chaos Labs architected such a system using LangChain and LangGraph.
Thanks for reading! Until next time, keep learning and building!
What did you think of today's issue? |
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