Hi all,

This week's newsletter will go out in two parts - there are too many links!

March has been busy over here, but more AGT events are coming - in the meantime, I hope to catch some of you at the other agent events in NYC.

Also: I'm launching an agent workshop series soon - AGT NYC members have priority (and free tickets) - let me know if you'd like to work together.

Cheers,
Ivan

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

Events

• Mar 15 - Build & Deploy Voice Agent Workshop + Hackathon
• Mar 18 - NYC AI Users - AI Talks, Demo & Social: AI Investing & Coding Agents
• Mar 18 - OpenClaw NYC Social
• Mar 18 - NYC AI Users - AI Talks, Demo & Social: AI Investing & Coding Agents
• Mar 18 - AI Engineering Happy Hour: Architecting Autonomous Agent Fleets
• Mar 19 - AI Builders Day NYC
• Mar 19 - Claw City NYC
• Mar 24 - LangChain Presents: It Worked on My Laptop - Agents in Production (New York)
• Mar 24 - OpenClaw Show and Tell Night
• Mar 25 - What Happens After You Actually Deploy AI Agents for CX?
• Mar 30 - Scaling up AI Agent Infrastructure

You can find more events on the AGT NYC Luma calendar.

News

• Microsoft taps Anthropic for Copilot Cowork in push for AI agents
• AI Agents Are Recruiting Humans To Observe The Offline World
• Google to Provide Pentagon With AI Agents for Unclassified Work
• Lobster buffet: China’s tech firms feast on OpenClaw as companies race to deploy AI agents
• China warns state-owned firms and government agencies against OpenClaw AI, sources say

OpenClaw mania and immediate government backlash aside, it's interesting that large organizations like Microsoft and the Pentagon aren't building agents in-house. The thinking must be that agent expertise is now sufficiently sophisticated that the leaders in the space have a meaningful advantage.

Launches

• Notion: Introducing Custom Agents
• Ramp Agent Card
• ZHC - AI-powered company builder
• OpenAI: Codex Security now in research preview
• Introducing OpenClaw on Amazon Lightsail to run your autonomous private AI agents
• Many agents, one team: Scaling modernization on Azure
• Google: Build dynamic agentic workflows in Opal

It seems like many hot startups from the 2010s are reinventing themselves around or aligning themselves with agents: Ramp, Notion, Airtable. The low-hanging fruit opportunities are drying up quick - the incumbents are figuring out how they fit into the new trend and are executing quickly.

Deals

• Temporal ($300M) - durable execution for agents
• Wonderful ($150M) - enterprise agent platform
• Onyx ($40M) - enterprise agent security
• Lyzr ($14M) - enterprise agent platform
• UnityAI ($8M) - healthcare workforce agents
• t54 Labs ($5M) - trust for agents
• Quotient AI acquired by Databricks

Agent platforms and infrastructure are dominating funding rounds this time around. I wonder how the first wave of these companies will fare long-term as the definition of "agent" evolves and ripples through the products built with previous assumptions.

Articles

• Don't trust AI agents by Gavriel Cohen
• Designing AI agents to resist prompt injection by OpenAI
• The Anatomy of an Agent Harness by Vivek Trivedy
• Expensively Quadratic: the LLM Agent Cost Curve by Philip Zeyliger
• How We Built Secure, Scalable Agent Sandbox Infrastructure by Larsen Cundric
• Where AI agents are heading: what we learned from recent YC startups by E2B
• Difference Between Agent Harnesses & Agent Frameworks by Tony Kipkemboi
• "SaaS is dead. Agents killed it." but also "SaaS is not going anywhere."
• You Need to Rewrite Your CLI for AI Agents by Justin Poehnelt
• Filesystems as the context layer for AI agents by Box
• Closing the verification loop: Observability-driven harnesses for building with agents by Datadog

The top agent trends so far in 2026 are harnesses, sandboxes, and file systems. Frameworks are starting to recede, along with anything that looks a little too hand-crafted. There also seems to be a trending focus on simplicity - whether it's advice to avoid overengineering harnesses or tiny OpenClaw alternatives, people want agents to be simpler.

Projects

• autoresearch - autonomous research agents
• Symphony - agent work management
• Paperclip - agent team orchestration
• Hermes Agent - self-improving agent
• Hermes Agent Self-Evolution - Hermes + DSPy + GEPA
• OpenFang - agent operating system
• Agent Browser Protocol - Chromium + MCP + REST

Long term, I see coding agents leading and merging with the rest of the agent space - they have the most experience, visibility, and practice. The thinking around coding agents now will be applicable to all future agents.

Learning

• Evaluating Skills
• Common workflow patterns for AI agents, and when to use them
• Local Agents with llama.cpp and Pi
• Agent Fundamentals
• Deploying AI Agents to Production: Architecture, Infrastructure, and Implementation Roadmap
• Building AI agents for 127 million customers: Practical lessons from Nubank [50:19]

More material is coming out to refresh the best practices for agents in 2026. No slowdown in sight yet - everyone is still learning what works.

Research

• Durable Execution for AI Agents: A Design Pattern for Fault-Tolerant Agent Loops
• OpenClaw-RL: Train Any Agent Simply by Talking
• CyberGym: Evaluating AI Agents' Real-World Cybersecurity Capabilities at Scale
• PostTrainBench: Can LLM Agents Automate LLM Post-Training?
• Tool-Genesis: A Task-Driven Tool Creation Benchmark for Self-Evolving Language Agent
• Agents of Chaos
• Ferret-UI Lite: Lessons from Building Small On-Device GUI Agents
• Benchmarking large language model-based agent systems for clinical decision tasks
• Towards a Science of AI Agent Reliability
• KARL: A Faster Agent for Enterprise Knowledge, powered by custom RL

Interesting reads all around. There has been a subtle shift recently in papers where the focus is more on practical and real-world use cases, which is a welcome change from the more hypothetical and theoretical papers published during 2023 and 2024.

Comments, suggestions? Reply to this email, let me know what you think!