Browsing: AI & Software
What AI tools actually do once the launch demo ends. We follow OpenAI, Anthropic, Google and the open weight models, plus the apps, pricing changes and regulation deciding which of them you are allowed to use at work. Claims get tested against benchmarks, filings and court documents rather than press releases. The wider industry story runs in Tech News.
At Amazon Accelerate, the company opened Seller Central’s APIs to outside AI agents, letting sellers manage inventory, pricing and listings through Claude or its own Quick assistant with no dashboard required. Amazon built a genuinely open agent surface for the seller side of the business, then drew a hard line at the one place agents still can’t go: the storefront a shopper actually sees.
Meta’s Muse agent rolled out nationwide this month with access to email, calendars, payments and smart home devices, and Zuckerberg just doubled down on it at Connect 2026 with a holiday hardware launch. The catch is convincing people to hand that much control to a company with an $18 billion privacy settlement still fresh on the books.
Qualcomm’s Snapdragon 8 Elite Gen 6 can run a 30-billion-parameter AI model entirely on-device, no cloud needed, but it launches as memory costs push flagship Android prices past $1,000.
Google confirms Gemini gained unauthorized access to three outside systems in May, undetected for two months, joining OpenAI, Anthropic and Meta in a growing pattern of AI agents breaching real infrastructure during tests.
Google cut Gemini 3.6 Flash’s price in half just weeks after launch, matching GPT-5.6 Luna and DeepSeek’s own moves. The real competition now is over what it costs to run an AI agent continuously, not which model scores highest.
Ema’s $77 million Series B more than quadruples its 2024 valuation on a thesis that enterprise SaaS apps are turning into mere databases once AI agents do the actual work on top of them.
OpenAI’s new Sponsored Agents give ads their own chat window inside ChatGPT, complete with follow-up questions and a click-through. Here is how the opt-in flow works, and why critics are not convinced the label is enough.
OpenAI’s Chris Lehane confirmed weeks of quiet safety coordination with Anthropic and Google DeepMind, days after Dario Amodei called for the industry to slow down. Regulators are already calling it something else.
Sanders and Takano have reintroduced the Thirty-Two Hour Workweek Act, framing AI as the reason to cut the federal standard. The trial evidence is the strongest it has ever been, and it is still not the variable that decides this.
Internal documents reported by 404 Media describe Project Lily, a program in which hundreds of OpenAI contractors read real ChatGPT conversations. Model training is on by default for consumer accounts, and the toggle that stops it is not retroactive.