Hello everyone,

Thanks for signing up - this meetup is just getting started!

The intro call last week was great, thanks to everybody who joined. Some stats on the group so far:

• Role: 35% engineers, 35% founders, 30% everybody else
• Agent experience: 70% have built an agent project
• Topics of interest: single and multi-agent systems, agents in production on AWS, optimizing codebases for agents, latest agent trends, async/ambient agents, evals, and context engineering

For the first AGT NYC newsletter, I thought it'd be useful to start with some interesting links to share, discuss, and build up a library for the group.

If you have any agent news, projects, posts, videos, or papers, please let me know and I'll include them in the next edition.

News

• Amazon announces Bedrock AgentCore and agent marketplace
• OpenAI launches ChatGPT agent to handle complex tasks
• Databricks announces Agent Bricks, automating evals and tuning
• GitLab launches Duo agent platform to automate DevOps tasks
• Alibaba launches Accio Agent to create product plans from concepts

It's interesting how different companies approach their bets on agents. Amazon wants to help engineers build, sell, and buy them. OpenAI wants to deploy a single-player consumer experience. Databricks is aiming to make building agents a turnkey process. GitLab and Alibaba are evolving agents in their own domain, focusing on vertical applications.

Recent fundraising: Tavily ($25M), AIUC ($15M), Noma ($100M)

Posts

• AI Agents have, so far, mostly been a dud by Gary Marcus
• Framework for developing safe and trustworthy agents by Anthropic
• We need a new ethics for a world of AI agents in Nature
• Agents are just workflows, really by Amplify Partners
• Six Principles for Production AI Agents by app.build

The hype around agents can be overwhelming. Sometimes it's good to get a sober take (or two) and zero in on realistic possibilities of new tech. I do like that criticisms from the disillusioned come out at the same time as practical advice from the builders, there's still a lot to learn.

Videos

Anthropic: Prompting for Agents

How many bananas can fit into a Rivian? Anthropic figures it out with a walkthrough of agent prompting best practices.

Andrew Ng: State of AI Agents

Fireside chat with Andrew Ng discussing building successful AI agents with the "Lego brick" approach, the good side of vibe coding, and voice AI apps.

HumanLayer: 12-Factor Agents: Patterns of reliable LLM applications

A quick, hype-free overview of what it takes to actually deliver production-grade LLM applications, beyond the "prompt + tools + loop" recipe.

MLOps: Everything Hard About Building AI Agents Today

A discussion about practical ways to measure agent performance in domains where achieving success is complicated.

Solomon Hykes: Containing Agent Chaos

The creator of Docker introduces Container Use, a way for agents to work in isolated, customizable environments without interfering with each other.

Projects

• Eigent - multi-agent desktop app built on CAMEL AI
• Deep Agents - general purpose deep research agent framework
• Omnara - access and monitor coding agents from anywhere
• Agent Reinforcement Trainer - help agents learn with GRPO
• WebAgent - web research agent suite from Alibaba

These projects capture some of the more exciting trends in agents: desktop multi-agent systems, open deep research, agent access outside of just one device, and integrating broader advances like GRPO into any AI application.

Papers

• Small Language Models are the Future of Agentic AI
• Design Patterns for Securing LLM Agents against Prompt Injections
• Agents of Change: Self-Evolving LLM Agents for Strategic Planning
• Efficient Agents: Building Effective Agents While Reducing Cost
• Embodied Web Agents: Bridging Physical-Digital Realms for Integrated Agent Intelligence

There are some cool ideas across all of these papers. The limitations of agents are coming into focus and I think they'll be useful guidance on productive deployment in the future. The idea of letting agents interact in a large simulation is also really interesting.

​If you know anybody else that would be interested in AGT NYC, please refer them to agtnyc.com to sign up.

Cheers,
Ivan