Hi all,

Planning is underway for the next AGT NYC event. The theme is "what makes successful AI projects work." The format includes a discussion panel and demos from presenters. If you'd like to participate as a panelist or presenter, or if you know someone who would be a great fit, let me know!

Events

• Jan 14 - Build Your Own AI Phone Agent
• Jan 20 - Agentic AI Workshop
• Jan 21 - Agentic AI Summit (online)
• Jan 27 - Agentic AI in the Trenches
• Jan 28 - NYC Voice AI Meetup: The State of Voice Agents
• Mar 9 - AgentCon New York

News

• Meta Buys AI Startup Manus for More Than $2 Billion
• OWASP Drops First AI Agent Risk List
• Major tech CEO sounds alarm on AI agents
• FunctionGemma: Bringing bespoke function calling to the edge
• NVIDIA Debuts Nemotron 3 Family of Open Models
• AWS DevOps Agent helps you accelerate incident response and improve system reliability

Incumbents roll out their own agents or buy generalists, new models well-suited for agent workloads keep arriving, and the risk profile of agents gets clearer.

Fundraising

• Port ($100M) - agent management portal
• Echo ($35M) - container security agents
• Sequence ($20M) - rev ops agents
• Navier ($5.6M) - hardware design agents
• Orq.ai ($5.4M) - agent deployment platform

This is an interesting mix of agent-oriented companies, all with different approaches of integrating agents into their strategy. There is no typical agent startup it seems.

Articles

• True agentic AI is years away - here's why and how we get there by Tiernan Ray
• AI grew up and got a job: Lessons from 2025 on agents and trust by Google Cloud
• The importance of Agent Harness in 2026 by Phil Schmid
• We removed 80% of our agent’s tools by Vercel
• Agent Orchestration is Not the Future by Matt Freeman
• AI agents are starting to eat SaaS by Martin Alderson
• Dynamic context discovery by Cursor
• A Great Year With Our 20+ AI Agents, But a Rough Week by Jason Lemkin
• AI agent-driven browser automation for enterprise workflow management by AWS

All good reads above, touching on context engineering, agent systems, the difficulty of managing agents, the need to simplify agents, and how early all of this still is, even though it's already producing real change.

Projects

• Agent Skills - skills for agents
• Vibium - browser automation for agents
• Google Agent Development Kit for TypeScript
• A2UI - agent-driven interfaces
• CUGA - configurable generalist agent
• AgentFS - filesystem for agents
• AgentField - Kubernetes for agents

Not much in the way of hot new agent frameworks lately, but several tools are emerging built for agents specifically. I think this points to the focus in the space shifting from theory to application. The most basic and obvious ideas are taken.

Learning

• Kaggle's AI Agents intensive course with Google
• Building a Research Agent with Gemini 3 + Deep Agents (7:49)
• The AI Agents Roadmap Nobody Is Teaching You
• Don't Build Agents, Build Skills Instead (16:21)
• Production-Grade Observability for AI Agents
• From Stateless Nightmares to Durable Agents (22:12)
• Real-World Agent Examples with Gemini 3
• Foundation: Introduction to LangChain - Python
• Proactive Agents (16:50)
• Nvidia’s NeMo Agent Toolkit: Making Agents Reliable
• 39C3 - AI Agent, AI Spy (40:31)

Tons of good material, especially for beginners. It feels like now is a good time to (re)learn agents - the stuff from last January seems almost antiquated.

Research

• State of Agent Engineering
• AI Agents Driving ROI: Real-world use cases in action
• Comparing AI Agents to Cybersecurity Professionals in Real-World Penetration Testing
• Model-First Reasoning LLM Agents: Reducing Hallucinations through Explicit Problem Modeling
• Beyond Ten Turns: Unlocking Long-Horizon Agentic Search with Large-Scale Asynchronous RL
• Memory in the Age of AI Agents
• Adaptation of Agentic AI

All useful reads, but one quote stuck with me from the Comparing AI Agents to Cybersecurity Professionals paper: "ARTEMIS placed second overall, discovering 9 valid vulnerabilities with an 82% valid submission rate and outperforming 9 of 10 human participants." What other domains will soon have similar (or better) outcomes?

​If you know anybody else that would be interested in AGT NYC, invite them to join at agtnyc.com.

Cheers,
Ivan