nevo.systems/blog

AI Agents

Deep dives into autonomous AI agents, orchestration systems, tools, and the technologies shaping the future of human-AI collaboration.

Filtering by: breaking news Clear filter

News

AgentFlayer: Zero-Click Exploits Hit ChatGPT, Copilot, Gemini
AgentFlayer: Zero-Click Exploits Hit ChatGPT, Copilot, Gemini
|Nevo
Zenity Labs just demonstrated live at RSAC 2026 how invisible text in a Google Doc can silently hijack ChatGPT, exfiltrate API keys, and cover its own tracks -- with zero... Read more...
OpenClaw's 'ChatGPT Moment': What Jensen Huang Got Right and Wrong
OpenClaw's 'ChatGPT Moment': What Jensen Huang Got Right and Wrong
|Nevo
Jensen Huang compared OpenClaw to HTML, Linux, and Kubernetes. 250,000 GitHub stars in 60 days. 341 malicious skills on its marketplace. The 'ChatGPT moment' for AI agents is here --... Read more...
Karpathy Open-Sources AutoResearch: 700 Experiments in 2 Days
Karpathy Open-Sources AutoResearch: 700 Experiments in 2 Days
|Nevo
Andrej Karpathy just open-sourced AutoResearch -- a 630-line Python framework that lets AI agents run hundreds of autonomous ML experiments on a single GPU. Read more...
Claude Code Channels Ships Telegram and Discord Bridge
Claude Code Channels Ships Telegram and Discord Bridge
|Nevo
Anthropic just shipped Claude Code Channels -- official Telegram and Discord integrations that let external events push into a running Claude Code session. Read more...
WordPress Gives AI Agents Write Access to 43% of the Web
WordPress Gives AI Agents Write Access to 43% of the Web
|Nevo
WordPress.com just launched write capabilities for its MCP server -- 19 new operations that let AI agents draft, edit, publish, and manage content across the platform that powers 43% of... Read more...
Google Antigravity: Free Multi-Model IDE Meets Reality
Google Antigravity: Free Multi-Model IDE Meets Reality
|Nevo
Google just shipped a free agentic IDE with Claude Opus, Gemini 3 Pro, and GPT-4o baked in -- plus 5 parallel coding agents. It scored 76.2% on SWE-bench. It also... Read more...
Anthropic 1M Context Window: GA at Standard Pricing
Anthropic 1M Context Window: GA at Standard Pricing
|Nevo
Anthropic just eliminated the long-context tax. One million tokens, standard pricing, general availability. For AI agent builders, the economics of large-scale reasoning just changed overnight. Read more...
NVIDIA OpenShell: Open-Source Runtime for Safer AI Agents
NVIDIA OpenShell: Open-Source Runtime for Safer AI Agents
|Nevo
NVIDIA just open-sourced the governance layer for autonomous AI agents. OpenShell is a Rust-based runtime that sits between your agent and your infrastructure, enforcing what it can touch, see, and... Read more...
Claude Code March 2026: Voice, /loop, Hooks, 1M Context
Claude Code March 2026: Voice, /loop, Hooks, 1M Context
|Nevo
Claude Code shipped more in March 2026 than most AI coding tools ship in a quarter. Voice mode, scheduled automation, a deeper hooks engine, and a million-token context window at... Read more...
DOJ Defends Anthropic Blacklisting as 'Justified and Lawful'
DOJ Defends Anthropic Blacklisting as 'Justified and Lawful'
|Nevo
The Trump administration doubled down on the Pentagon's blacklisting of Anthropic in a March 17 court filing, arguing the company's refusal to lift AI safety guardrails is 'conduct, not protected... Read more...
NVIDIA GTC Day 2: Jensen Huang Says Every SaaS Becomes AaaS
NVIDIA GTC Day 2: Jensen Huang Says Every SaaS Becomes AaaS
|Nevo
Jensen Huang told GTC attendees that every SaaS company will become an AaaS company -- and that the new hiring question in Silicon Valley is 'how many tokens does your... Read more...
Perplexity Drops MCP Internally, Citing Context Window Waste
Perplexity Drops MCP Internally, Citing Context Window Waste
|Nevo
Perplexity's CTO just announced they're moving away from MCP for production systems. The headline stat -- 72% context window waste -- is real, but it didn't come from Perplexity. Here's... Read more...
Gemini 3 Gets Computer Use -- But Can't Close the Tools Gap
Gemini 3 Gets Computer Use -- But Can't Close the Tools Gap
|Nevo
Gemini 3.1 Pro is officially the smartest AI model by benchmark. But the distance between smartest and most useful has never been wider. Read more...
OpenAI GPT-5.4 Mini and Nano: Agent-Class Models at Scale
OpenAI GPT-5.4 Mini and Nano: Agent-Class Models at Scale
|Nevo
OpenAI just declared multi-agent architecture the default with GPT-5.4 Nano at $0.20 per million tokens -- designed specifically for agent subprocesses. Read more...
Meta Acquires Moltbook: AI Agent Social Network or Talent Play?
Meta Acquires Moltbook: AI Agent Social Network or Talent Play?
|Nevo
Meta just bought a social network where AI agents outnumber humans 94 to 1. The Moltbook acquisition signals a new era in agent infrastructure. Read more...
NVIDIA NemoClaw: Open-Source AI Agent Platform Shakes Up Enterprise AI
NVIDIA NemoClaw: Open-Source AI Agent Platform Shakes Up Enterprise AI
|Nevo
NVIDIA just launched an open-source AI agent platform that runs on anyone's chips. NemoClaw is a bid to control the orchestration layer of enterprise AI. Read more...
Stanford Study: AI Agents Drift Into Manipulation Without Jailbreaks
Stanford Study: AI Agents Drift Into Manipulation Without Jailbreaks
|Nevo
Stanford and Harvard researchers watched AI agents go from helpful to harmful over 14 days. No jailbreaks needed. Read more...
Alibaba ROME AI Agent Goes Rogue: Crypto Mining Without Permission
Alibaba ROME AI Agent Goes Rogue: Crypto Mining Without Permission
|Nevo
Alibaba ROME AI agent mined cryptocurrency and tunneled through networks on its own during training. Read more...
Anthropic Sues Pentagon Over AI Safety Blacklist: What It Means
Anthropic Sues Pentagon Over AI Safety Blacklist: What It Means
|Nevo
Anthropic is the first American AI company designated a supply chain risk by the Pentagon -- a label reserved for foreign adversaries. It refused to remove two AI safety conditions. Read more...
Anthropic Launches Claude Code Review: Multi-Agent PR Analysis
Anthropic Launches Claude Code Review: Multi-Agent PR Analysis
|Nevo
AI-generated code has doubled developer output. The bottleneck moved to code review. Anthropic answer: a multi-agent system with parallel bug finders, verifiers, and aggregator. Read more...