Language app Speak cut the wait between actions from 1.4 seconds to 0.13 seconds with OpenAI's new Decisions API, which lets an AI agent quickly pick from a few set options, like which team gets a support request. How to automate support with agents: https://t.co/izqVO6s9T4
@ArthurLiu0725 Every tool call. The agent pauses each time it calls a tool and wakes up when the result comes back, whether that takes a second or three days.
AWS and UC Berkeley released Infinity, open-source software that pauses an AI agent while it waits and wakes it when results arrive. AWS says it fits 75,000 agents running at once in the 8 GB of memory on a Raspberry Pi 5. How to run AI agents at home: https://t.co/Q5O6Kn1YJX
OpenAI added scheduled automations to its Agents API today. Agents can now run on a recurring schedule, like every morning at 10 AM, or once at a set time, and can schedule their own follow-ups.
OpenAI, PerturbAI and Scripps Research used close to 2,000 AI agents, one per gene, to analyze a whole-brain mouse CRISPR atlas in 32 minutes. They estimate it would take months to a year by hand.
How to split research across sub-agents: https://t.co/r0ihJsC1Vv
Walmart launched Agent Gateway, an MCP server letting approved AI agents like ChatGPT and Gemini shop Walmart and Sam's Club for customers. Agent Profiles with spending caps roll out next month. Amazon is blocking these agents.
Anthropic donated the Agent Skills open standard to the Agentic AI Foundation, which also hosts MCP. Anthropic says 1 million skills have been published and nearly 280 million installed.
How to scan a skill before you install it: https://t.co/6eSV5yegb3
Anthropic opened a research preview of Claude Duet, an app where Claude works as a proactive agent: it flags what needs you, keeps working while you're away, and asks before acting. Waitlist for US Pro and Max users.
Run a similar daily agent now. Li k in comments
PhoneUse launched today: an MCP server and API that gives AI agents their own phones, cloud Androids or real iPhones, to use native apps like Uber and WhatsApp.
A DIY version built with Codex and an iPhone: https://t.co/AyuTURTrL0
NVIDIA unveiled the OpenShell Reference Architecture at GTC Berlin: an open-source runtime that sandboxes AI agents, checks outbound requests against policy, and can quarantine an agent in milliseconds.
Self-host a sandbox for your own agent: https://t.co/ghlPzz8MEN
Asymptote Labs raised a $5.5M seed and launched Beacon, an open-source memory layer for coding agents. It captures sessions across Claude Code, Cursor, Codex and 20+ other tools and turns fixes that worked into reusable skills.
Shared agent memory setup: https://t.co/lnoINouQrP
Delta, JPMorgan, Schwab and Yelp have confirmed limits on outside AI agents booking flights or accessing customer accounts. Delta says no integration lets agents book its flights.
How one person had Claude Code book a cheap flight and chase the refund: https://t.co/Vs3NLrFyFr
NVIDIA App 11.0.9.565 adds a local MCP server that lets AI agents launch games, change graphics presets and start recordings. Users found it on after updating. Off switch: Settings > Features > Connect Agents & Apps.
A game about agent permissions: https://t.co/8NlIpI1O4m
India's CERT-In issued a critical advisory: AI agents under test reached live systems at Indian organizations and read confidential records using exposed credentials and loose cloud roles. No affected org detected it.
Scan your app for exposed keys: https://t.co/YrgJDtTfc1
While the internet argues over whether vibe coding is real, one builder had Claude Opus rebuild SimCity 2000 as a low-poly WebGL city. A few days later, it was playable
The particulars: https://t.co/mBQ75gijT6