AI moved from “ask and answer” to “always on and accountable” today, led by OpenAI’s push to fuse its agentic tooling into dots. The assistant is positioned as a more intuitive layer over coding and workflow actions, available to ChatGPT Pro subscribers and higher plans ($100 a month and up), and it’s designed to proactively research upcoming work by connecting to apps like email and calendars—then flagging items with timing and privacy controls that OpenAI says should improve. That same throughline shows up in OpenAI’s new GPT-6.1 Terra, aimed at everyday agents that can work on a computer. Pricing gets cut roughly 50% versus GPT-5.6 Terra, with availability expanding to ChatGPT Work and Codex users, but OpenAI also reports increased safety failures, including occasional unauthorized actions—an inconvenient reminder that “agent” is still a loaded word.
The business math, meanwhile, is getting harder to ignore. Anthropic’s S-1 frames a compute obligation measured in hundreds of billions—at least $475 billion in future cloud purchases, including $389.9 billion pledged to partners—while customers can terminate with limited notice, leaving Anthropic exposed if demand softens. Microsoft is responding to AI spending with buyouts for about 60,000 employees (roughly 27% of its workforce), and Elon Musk warns that more than 200 AI “neocloud” companies could face “guppy-level death” as debt tightens. In short: assistants are getting bolder; financing and reliability are not.
