AI Agents Weekly

March 23, 2026 — Your weekly dose of AI agent news

🤖 AI Agents Weekly

March 23, 2026

Opening

The race to build fully autonomous AI researchers has shifted into high gear. This week, both a high-profile demo and a major corporate pivot showed that the era of AI agents conducting their own scientific discovery is no longer a distant theory—it's an active engineering sprint.

Top Stories

OpenAI Pivots to Build a Fully Automated AI Researcher
MIT Tech Review reports that OpenAI is refocusing its research efforts and resources on a "grand challenge": building an AI that can autonomously conduct end-to-end scientific research. This isn't just a side project; it's a major strategic shift for one of the world's leading AI labs, signaling where they believe the next breakthrough will come from.
Read more →

Karpathy's Autonomous Agent Runs 700 Experiments in 48 Hours
In a stunning demonstration of what's possible now, Andrej Karpathy shared results from an autonomous AI research agent that designed and ran 700 distinct experiments over a single weekend. This practical glimpse shows the staggering speed and scale at which AI-driven discovery can already operate, even outside giant corporate labs.
Read more →

Gemini's Task Automation: Clunky But Impressive
The Verge's hands-on with Gemini's new feature that lets the AI use apps on your phone (like ordering Uber or DoorDash) reveals a familiar pattern: the first-generation product is slow and limited, but the core capability—an agent taking actions in the real digital world—is undeniably powerful and points to a hands-free future.
Read more →

WordPress.com Opens the Gates to AI Publishing Agents
WordPress.com now natively allows AI agents to write, manage, and publish posts. This massively lowers the barrier for automated content creation but also sets the stage for a significant increase in machine-generated material across the web, forcing a rethink of content authenticity and SEO.
Read more →

Major Legal Consolidation Against OpenAI Over Chatbot Harm
Over a dozen lawsuits in California alleging harm (including suicide) linked to ChatGPT interactions have been consolidated into one major litigation. This represents a critical legal front for the AI industry, testing liability for the actions and outputs of conversational agents.
Read more →

Quick Hits

• Nvidia GTC Roundup: Jensen Huang's keynote highlighted NeMoClaw, Robot Olaf, and a projected $1 trillion data center bet underpinning the AI infrastructure boom. TechCrunch
• Efficiency Breakthrough: Flash-MoE project demonstrates how to run a massive 397B parameter model on a standard laptop. GitHub
• Developer Deep-Dive: How "context engineering" can turn Codex into a full dev team while drastically cutting token waste and cost. Reddit
• Tool of the Week: "Agent Kernel" – a simple, open-source method using three Markdown files to give any AI agent persistent memory and state. GitHub

Closing

This week perfectly captured the dichotomy of AI agents: breathtaking demonstrations of autonomous capability running alongside real-world growing pains—clunky UX, legal challenges, and content floods. The trajectory, however, is unmistakable. The agents are moving from the lab to the wild.

Until next week,
The Editor
AI Agents Weekly

Want to change how you receive these emails?
You can update your preferences or unsubscribe.

Curated by Paxrel — Powered by AI, reviewed by humans.

Was this forwarded to you? Subscribe here