Hi all,

We've managed to leave the NYC snowpocalypse mostly unscathed - time for a meetup soon.

Making some predictions for this year, before everything changes again:

1. Claude Code will emerge as the first popular AI agent platform, with competition on the professional side from Big Tech and frontier labs and from the developer side with tools like OpenCode and Kilo Code
2. Skills and CLIs will win over MCP for extending agent capabilities
3. Non-technical people will start using Claude Code (and Cowork)
4. Websites and web apps (e.g. SaaS) without an agent-friendly interface will start losing ground to experiences that optimize agent experience

Reminder: for the next AGT event, we're looking for presenters and panelists to discuss and demo successful AI projects and what it takes to make them work. If you'd like to get involved, please fill out this form.

Events

• Feb 17 - NYC: LangChain Presents: Agent Observability Powers Agent Evaluation
• Feb 17 - Demo Night: OpenClaw
• Feb 18 - Dinner: Leveraging AI Agents
• Feb 19 - Agent Builder & Elasticsearch Results with LambdaMART
• Feb 21 - Self Improving Agents Hack - NYC
• Feb 23 - Build & Deploy Voice Agent Workshop + Hackathon (Vapi AI x Cartesia)
• Feb 26 - CTOs at Scale (NYC): Agentic Experiments
• Mar 9 - AgentCon New York

News

• Ramp launches accounting AI agent
• OpenAI introduces Frontier, a platform to deploy and manage AI agents
• Are AI agents ready for the workplace? A new benchmark raises doubts
• AI agents now have their own Reddit-style social network, and it’s getting weird fast
• AI agents are posting gigs for humans, and some of them are very weird
• Cadence introduces an AI agent to speed up computer chip design
• Multi-agent AI and robots automate materials discovery in closed-loop lab system
• GitHub introduces the Agents tab in repositories
• AI Agents Are Here to Stay, Businesses Say
• Introducing AI Cyber Model Arena: A Real-World Benchmark for AI Agents in Cybersecurity

Between Moltbook, agents recruiting humans, and OpenAI Frontier, many new agent developments have hit the market in the last few weeks. There's a real sense that a lot of work from last year is just now starting to appear in finished form.

Deals

• Resolve AI ($125M) - agents for production
• Bretton AI ($75M) - financial compliance agents
• Fieldguide ($75M) - agents for audits
• Entire.io ($60M) - agent developer platform
• Fiddler ($30M) - agent control plane
• Daytona ($24M) - computers for agents
• Sapiom ($15M) - agent payment system
• Synthpop ($15M) - healthcare workflow agents
• EnFi ($15M) - credit analyst agents
• Take2 ($14M) - healthcare hiring agents
• Concourse ($12M) - financial stack agents
• Veritus ($10M) - consumer lender agents
• Polybee ($4M) - physical pollination agents
• Pallma AI ($1.6M) - AI agent security
• Tavily acquired by Nebius

Recent funding rounds are geared heavily towards finance, compliance, and healthcare. Agent infrastructure and ecosystem investments are happening, but it's clear the focus is on applications in legacy domains.

Articles

• Lessons from Building AI Agents for Financial Services by Nicolas Bustamante
• What I learned building an opinionated and minimal coding agent by Mario Zechner
• Lessons From 2 Billion Agentic Workflows by Joao Moura
• The Bitter Lesson of Agent Frameworks by Gregor Zunic
• I Let Google’s ‘Auto Browse’ AI Agent Take Over Chrome. It Didn’t Quite Click in WIRED
• The emerging agent architecture by Arvind Jain
• I Loved My OpenClaw AI Agent—Until It Turned on Me in WIRED
• The Agentic Enterprise in 2026 by Mayfield
• The Math on AI Agents Doesn’t Add Up in WIRED
• I spent 6 hours in Moltbook. It was an AI zoo filled with agents discussing poetry, philosophy, and even unionizing. by Henry Chandonnet
• From runtime risk to real‑time defense: Securing AI agents by Microsoft Defender Security Research Team
• Keeping your data safe when an AI agent clicks a link by OpenAI
• A Missing Layer in Agentic Systems by Joao Moura
• The two patterns by which agents connect sandboxes by LangChain
• You could've invented OpenClaw by Nader Dabit
• Is Your Workplace Set Up for AI Agents? in HBR
• Inside OpenAI’s in-house data agent by OpenAI authors

Several reflections and lessons learned have been published recently, summarizing the rocky progress of AI agent technology last year. Reading between the lines, things are getting more sophisticated fast - patterns from a couple of years ago are long obsolete.

Projects

• OpenClaw - personal AI assistant
• Pi - AI agent toolkit
• Tapes - transparent agent telemetry
• Terminal Use - background agent orchestration
• Dexter - financial research agent
• Moltworker - OpenClaw on Cloudflare
• Hive - outcome driven agents
• Agentic Context Engine - agent feedback learning
• BoxLite - sandbox for agents
• Agent SOP - natural language workflows

OpenClaw has absolutely exploded recently as one of the fastest-growing GitHub repositories of all time. In the meantime, other projects that are helping build agent systems keep on coming.

Learning

• Agent Skills with Anthropic
• AGENTS.md outperforms skills in our agent evals
• Build AI agents with Amazon Bedrock AgentCore using AWS CloudFormation
• The Complete Guide to Building Skills for Claude
• Shell + Skills + Compaction: Tips for long-running agents that do real work

This list is skills heavy, though I think it's only because the tech is still fairly new. Users will figure out how to use them soon and they'll fade into the background.

Research

• Agentic Reasoning for Large Language Models
• AOrchestra: Automating Sub-Agent Creation for Agentic Orchestration
• Agent Primitives: Reusable Latent Building Blocks for Multi-Agent Systems
• SkillRL: Evolving Agents via Recursive Skill-Augmented Reinforcement Learning
• Helping AI agents search to get the best results out of large language models

Some mind-bending stuff here, especially around automatic subagent creation. If agents are programs, then we're getting programs that write their own programs. Really interesting and exciting to see research move in this direction - it's not theoretical after all.

Invite a friend to join AGT NYC at agtnyc.com!

Cheers,
Ivan