22 articles tagged “ai-agents”, most recent first.
n8n’s guide explains how autonomous agents work, where they help, and why tool access makes governance essential.
A new n8n guide explains when AI builders should use predictable APIs, dynamic MCP tools, or both together.
Anthropic’s latest threat report says AI is helping less-skilled actors run faster, broader campaigns across seven areas of harm.
GPT-6 Astra brings computer use, coding, browsing and document reasoning to existing business workflows, with API pricing from $10 per million input tokens.
n8n’s new assistant builds, tests, debugs, and hands over editable workflows while keeping users in control of credentials and activation.
n8n’s new guide breaks agent reliability into controls, debugging, evaluation, metrics, and ongoing monitoring.
An MIT researcher connected Codex to lab software, letting GPT-5.6 Sol run and refine routine quantum-chip measurements with limited supervision.
OpenAI outlines how GPT-6 Astra, agents and custom chips could make more complex AI workflows practical at scale.
OpenAI reports sharp growth in internal coding-agent use while stressing that human oversight and safety remain essential.
GPT-6 Astra targets autonomous computer work, software engineering, science, and safer delegation with major benchmark gains.
Anthropic’s Claude Fable 5.1 targets agent builders with four-times-cheaper cache reads and stronger terminal benchmarks.
Google is giving selected governments and enterprises access to AI tools designed to find, verify, and patch vulnerabilities at scale.
A replay pipeline tests model replacements against real agent conditions, revealing why safety gates should not be buried in a single score.
Basis, Clay, and Exa show how persistent context, tools, and human review can move agents from assistance to repeatable execution.
Long-running agents need durable state, controlled context, and deterministic workflows to avoid drift, corruption, and hallucinations.
As coding agents make production easy, Hugging Face argues that human editorial judgment is now the critical skill.
The three.ws project combines generative 3D avatars, agent memory, tool use and autonomous USDC payments across several developer surfaces.
Cloudflare’s BotBase update lets AI and web-bot operators track reviews, fix listings, and explain how their crawlers use content.
n8n argues that static permissions cannot keep pace with autonomous agents and recommends real-time, task-scoped enforcement.
OpenAI calls a July 2026 security incident a warning that capable AI agents can evade isolation and coordinate without permission.
NVIDIA’s open 30B MoE targets the repetitive tool calls that make long-running AI agents slow and expensive.
A token-handling mistake in multi-turn RL can silently corrupt gradients when models call tools mid-rollout.
Meshio