Gemini 2.0 and Google's new agents

Plus: Amazon launches an AGI lab, the secret to agentic process design, and more

Author

Michael Cunningham
December 12, 2024

In partnership with

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

With Building AI Agents’ writing staff on strike, please enjoy this compilation of our greatest hits! Just kidding, this newsletter is always human-written, so if it’s the only thing in your inbox this morning…you know why

In today’s issue…

  • Google releases Gemini 2.0 with new agents

  • Amazon’s AGI lab

  • How agents will transform workflows—while keeping humans central

  • Testing web browser agents with BrowserGym

…and more

📰 NEWS

Source: Google

Google unveiled its new Gemini 2.0 model with a heavy emphasis on agentic applications, touting new features such as multimodal reasoning, complex instruction following, multi-turn function calling, and more. To drive home the model's agentic focus, the company simultaneously showcased improvements to its Project Astra agent and released a new web-browsing agent called Project Mariner and a coding agent called Jules.

Amazon announced the formation of its ambitiously-named AGI SF lab, a small, elite team devoted to building advanced AI agents capable of taking actions in the real world.

Riding high on the success of its software engineering Replit Agent, the eponymous startup launched Replit Assistant, an agentic application designed to refine and refactor existing projects to suit users’ instructions.

The LLM builder hosted over 100 developers in San Francisco for a hackathon to promote its new open standard for agent communication, the Model Context Protocol.

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

Refer one person and we’ll send you a set of weekly bonus links to additional AI agent news for a month! See here for a sample

🤝 WITH WRITER

Writer RAG tool: build production-ready RAG apps in minutes

  • Writer RAG Tool: build production-ready RAG apps in minutes with simple API calls.

  • Knowledge Graph integration for intelligent data retrieval and AI-powered interactions.

  • Streamlined full-stack platform eliminates complex setups for scalable, accurate AI workflows.

🛠️ USEFUL STUFF

Source: Lamini AI

Categorization Agent Toolkit (CAT) is an SDK by agent startup Lamini AI which enables users to build agents for intelligently classifying textual data for applications such as sentiment analysis and request urgency.

Open-source agent framework Phidata—featured in the very first issue of Building AI Agents—has released a major slate of new features, including support for multimodal agents, a new method for building agent pipelines, and more.

💡 ANALYSIS

Japanese characters for kaizen | Source: Wikimedia Commons

This Harvard Business Review article compares the new era of agentic automation to the revolutionary “kaizen” style of manufacturing pioneered by Japanese auto manufacturers. In the authors’ reckoning, agents will enable a new “kaizen 2.0” which will again reinvent companies’ workflows—but maintain the original’s focus on human empowerment.

A look into the hot competition among tech companies to capture the E-commerce agent space, including the author’s (somewhat shaky) experience with Perplexity’s shopping agent.

This piece by the consulting firm charts the move from large, general-purpose LLMs to smaller, more specialized ones—and the paradigm shift they will induce by powering broadly capable AI agents.

The author of this piece reviews the three major trends that will shape AI in 2025: multimodality, small language models, and agents.

🧪 RESEARCH

Created by the author using Dall-E 3

With the explosive proliferation of web browser agents in the past several months, there is a growing need for a common evaluation methodology. BrowserGym is a new unified environment for testing agents’ capabilities on a wide range of web tasks.

This paper introduces a new agentic system for creating synthetic customer interaction data, allowing companies to improve their agents’ capabilities on actual customer data by fine-tuning on the outputs.

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