Hi all,

Got another roundup of agent links for you. Even though August has been a slow month, there's still a lot of things happening in the agent space.

Gartner's updated hype cycle earlier this month shows AI agents at the peak of inflated expectations. I think that's about right, because agent builders know that the hype is a bit ahead of reality. Gartner also suggests that 40% of apps will add AI agents by 2026 and that leaders need to act fast. I guess we're in for a lot of hype-driven development.

The first in-person event is still being planned, considering late September. Any referrals to a venue or event sponsor would be greatly appreciated.

If you have any agent news, projects, posts, videos, or papers, let me know!

News

• Anthropic launched Claude for Chrome
• Ai2 launches Asta, a standard for agents in science
• Salesforce launches CRMArena-Pro to stress test AI agents
• Cloudflare and Browserbase launch Web Bot Auth standard
• AI agents have a long way to go

The agent standards keep coming - AGENTS.md earlier this month and now new proposals for identity and science. There's clearly a lot of work being done to understand how to integrate otherwise chaotic agents into existing complex workflows and systems. It still feels really early, like the very beginning of a new industry.

Recent fundraising: Maisa ($25M), Archestra ($3M), Bluejay ($4M)

Posts

• Why Everything's an "AI Agent" Now in New York Magazine
• Don't Built Multi-Agents by Cognition
• Hidden Costs of Agentic AI by Galileo
• Scaling Agents Beyond Token Limits by Factory.ai
• The Rise of Computer Use and Agentic Coworkers by a16z

Agents in production don't act like agents in theory. There's still a lot to learn about how to properly engineer and operate these systems. I think the limits of existing tech will be firmly understood over the next year or so, maybe prompting the next generation of software for building agents.

Videos

OpenPipe: Building Reliable Agents with RL

Kyle Corbitt talks about using GRPO to help agents learn from successes and failures. Learn from case studies and real world experience about using RL for agents.

CB Insights: What Powers the Smartest AI Agents

Analysis from CB Insights on the current AI agent stacks deployed in production. Key enterprise use cases and challenges are also discussed.

Human Layer: Advanced Context Engineering for Agents

Dexter Horthy shares what Human Layer has learned about scaling coding agents in real world projects. The session, recorded at YC Root Access, dives into optimizing context for agents and the benefits of spec-first development.

Palo Alto Networks: Breaking AI Agents

Jay Chen discusses security issues in AWS Bedrock Agents, demonstrating how attackers can exploit prompt injection and misuse tools to compromise agents. Broader implications are discussed, along with mitigation strategies.

Sayash Kapoor: Building and evaluating AI Agents

A broader talk from earlier this year covering the limitations of AI agents and how to use evals to improve performance. Worth a watch to see what still applies four months later.

Projects

• Julep - Firebase for AI agents
• Cloudflare Agents - agent framework for Cloudflare
• Graphiti - real-time knowledge graphs for agents
• Rowboat - build AI agents with natural language
• PocketFlow - let agents build agents

The whole agents space feels like it's on fast-forward when it comes to experimentation and integration. Mind-bending concepts, like letting agents build agents, are regular everyday projects. It's also good to see that challenging problems, like keeping knowledge bases up to date, actually have promising attempted solutions.

Papers

• Exploring Autonomous Agents: A Closer Look at Why They Fail When Completing Tasks
• FutureX: An Advanced Live Benchmark for LLM Agents in Future Prediction
• OpenCUA: Open Foundations for Computer-Use Agents
• A Comprehensive Review of AI Agents: Transforming Possibilities in Technology and Beyond
• When AIs Judge AIs: The Rise of Agent-as-a-Judge Evaluation for LLMs

Agent papers keep getting published even in the slow summer days. It's interesting to see research that goes beyond using agents just to complete business tasks, there may be a lot of value to uncover in this direction. Agents could be useful in ways we haven't even considered yet.

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

Cheers,
Ivan