ChatGPT Voice hits the desktop and steers Codex by voice
Full-duplex GPT-Live now drives your computer and multiple agents from one command. Also: Hugging Face's leaner MCP, and a US bill to unmask AI crawlers.
Copy markdown
One spoken command, a full PR
OpenAI began rolling ChatGPT Voice to the macOS and Windows desktop apps on July 24, powered by the full-duplex GPT-Live model that speaks, listens, and works at once. In OpenAI's demo, a single command opened a thread, filed a pull request, and traced a bug's root cause across ChatGPT Work and Codex agents; on macOS it reads your screen via Appshots and browses with computer-use skills.
Hugging Face folds its MCP into one tool
Hugging Face's MCP Server overhaul swaps tool sprawl for a single, search-equipped hf_fs that reaches every repo, bucket, doc, and paper in about 1,000 tokens, so agents spend far less context on the Hub. New Sandboxes bolt secure execution onto your repos and buckets for dataset analysis, training, or spinning up a Space — enough to point an agent at Hugging Face this weekend.
A US bill moves to unmask AI crawlers
A bipartisan federal bill introduced July 24 would require AI web crawlers to disclose their identity and purpose, targeting stealth bots that now exceed half of traffic on some sites. If it passes, undisclosed scraping — for training or agent browsing — turns legally risky, so wire an honest user-agent into your crawlers now.