Hi all,

We had a great breakfast event yesterday, co-hosted with Topher Price and Will McKelvey at the beautiful Lerer Hippeau offices. Thank you to everybody who joined and I hope to see you all at the next event in a few weeks!

The "year of AI agents" is coming to a close, and even though we all haven't lost our jobs yet to agents, the agent trend has become massive and unstoppable. The hype cycle seems to be progressing towards the long plateau of productivity and it may take us at least the rest of the decade to get there.

The new AGT NYC site is almost ready and will be launched in January. With the new site, you'll be able to go through all the content from past newsletters, follow the latest news and events, and learn more about building agents. Get ready for it to launch soon!

Events

• Agentic AI Summit - 1/21/26, online
• AI Agent Event - 2/21/26, Fort Lauderdale
• NVIDIA GTC AI Conference - 3/16/26, San Diego
• AI Agent & Copilot Summit - 3/17/26, San Diego
• AGENT 2026 - 4/12/26, Rio de Janeiro
• AI Agent Conference 2026 - 5/4/26, NYC
• Agentic & Applied AI for the Enterprise Conference - 6/8/26, Atlanta
• Ai4 - 8/4/26, Las Vegas
• Oracle AI World - 10/26/26, Las Vegas
• AGENTICS 2026 - 10/28/26, France

News

• Linux Foundation Announces the Formation of the Agentic AI Foundation (AAIF)
• Gartner predicts AI Agents will outnumber sellers by 10X and 90% of B2B buying will be AI agent intermediated
• Introducing Google Workspace Studio: Automate everyday work with AI agents
• AWS unveils frontier agents, a new class of AI agents that work as an extension of your software development team
• AI Agents Break Rules Under Everyday Pressure
• Introducing Lux, the World's Best Foundation Computer-Use Model
• Google launches managed MCP servers that let AI agents simply plug into its tools

Big Tech companies continue to wheel out their agent offerings, aiming at higher levels than frameworks and SDKs. It's probably not a coincidence that a foundation is being stood up right now to contain the sprawl of solutions and standards being produced in the industry. It will be interesting to see what will remain relevant in the foundation a year from now.

Fundraising

• Archetype AI ($35M) - physical agent deployment
• Shapes ($24M) - agents for HR
• Simular ($21.5M) - computer use agents
• Prime Security ($20M) - agentic product security
• Vijil ($17M) - agent resiliency platform
• Raindrop ($15M) - agent monitoring platform
• Cleric ($9.8M) - agents for SRE
• Orq.ai ($5.4M) - agent deployment platform

Exciting to see so many startups take a swing at reinventing conventional developer tools for the agent era. I wonder how many of them will survive by themselves, however, instead of being outcompeted or acquired by the incumbents. Also I'm wary of the same fragmentation seen in agent frameworks starting to appear at the product and business level - waves of consolidation ahead seem highly likely now.

Articles

• Why (Senior) Engineers Struggle to Build AI Agents by Phil Schmid
• AI Agents Aren’t Ready for Consumer-Facing Work — But They Can Excel at Internal Processes in HBR
• Building the agent economy: How cloud leaders are shaping AI’s next frontier by CB Insights
• Why AI agents failed to take over in 2025 in ZDNET
• Agents should be more opinionated by Vivek Trivedy
• How GitHub’s agentic security principles make our AI agents as secure as possible
• Deepnote CEO: Why Notebooks Are the ‘Perfect User Interface’ for AI Agents in The New Stack
• Microsoft’s Attempts to Sell AI Agents Are Turning Into a Disaster in Futurism
• Silicon Valley Builds Amazon and Gmail Copycats to Train A.I. Agents in NYT
• 8 learnings from 1 year of agents by PostHog
• Goodbye Dashboards: Agents Deliver Answers, Not Just Reports in The New Stack
• AI Agents in Action: Foundations for Evaluation and Governance by WEF
• HBR: Only 6% of companies fully trust AI agents to handle core business processes in Fortune

I think it's safe to say we're still figuring agents out. The limitations of current approaches don't seem easy to overcome. Building agents may turn out to be a mix of art and science, just like building software. That said, it's likely we'll start figuring out where agents are a good fit sooner rather than later. It's clear they're not a universal solution yet.

Projects

• AG-UI - Agent-User Interaction Protocol
• Better Agents - standards for agent building
• Acontext - context data platform for agents
• Raptor - Claude Code for security
• AI Agents From Scratch - local agents without frameworks

Interesting to see developer focus shift away from frameworks to higher-level targets. Protocols and standards are becoming necessary, along with infrastructure that agents need to do their job. The work emerging around converting coding assistants into special-purpose agents is also really intriguing - there may not be a need to create bespoke agents for certain jobs after all.

Learning

• Agentic AI with Andrew Ng
• Context Engineering for AI Agents: Lessons from Building Manus
• Effective harnesses for long-running agents
• Introducing Strands Agent SOPs - Natural Language Workflows for AI Agents
• Katelyn Lesse – Evolving Claude APIs for Agents, Anthropic (13:24)
• Of course you can build dynamic AI agents with Temporal
• AI Agents in Production: Lessons from Rippling and LangChain (43:38)
• Build reliable AI agents for UI workflow automation with Amazon Nova Act
• Train Small Orchestration Agents to Solve Big Problems
• Building Coding Agents with Tool Execution with E2B

Agents have been around long enough that the latest learning material is accumulating some high-quality lessons. At this point I skip anything made before 2025 as it's already obsolete. The space moves fast, but the latest ways tend to be simpler and more powerful anyway.

Research

• Measuring Agents in Production
• The Adoption and Usage of AI Agents: Early Evidence from Perplexity
• State of AI - An Empirical 100 Trillion Token Study with OpenRouter
• Agent Lightning: Adding reinforcement learning to AI agents without code rewrites
• Fara-7B: An Efficient Agentic Model for Computer Use
• Towards a Science of Scaling Agent Systems
• Comparing AI Agents to Cybersecurity Professionals in Real-World Penetration Testing
• Agent0: Unleashing Self-Evolving Agents from Zero Data via Tool-Integrated Reasoning

One interesting theme in the materials above is the idea of enabling improvements in agents without additional work. Even though data is cheap and code is nearly free, the efforts continue to build the simplest possible agent that can bootstrap into a much more powerful version of itself. We may find there's not all that much complexity to the ideal agent architecture.

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

Cheers,
Ivan