AI-generated illustration
Today’s AI roundup highlights significant updates across cloud development sessions, workplace software integrations, and efficient on-device computing hardware. Tech companies continue expanding how developer tools and artificial intelligence agents function across daily workflows in the apps sector.
Major AI roundup Developments
Anthropic announced that Claude Code cloud sessions have exited research preview to become officially available. This update enables tasks to persist even after users close their laptops, and Anthropic is providing a $100 credit for Pro subscribers alongside $250 for Max users, as reported by @ClaudeDevs.
In addition, Anthropic launched Claude Marketplace as a single directory hosting integrations, consulting partners, and agent purchases. The directory includes connectors for Slack and Notion, software from Cursor and CrowdStrike, and consulting services from Accenture and Deloitte, according to @claudeai.
OpenAI updated ChatGPT Voice inside ChatGPT Work on web and mobile devices to access plugins for Slack, email, and calendars. Furthermore, the voice feature can now run on Astra, Sol, or Luna models to support professional users in the modern economy, according to @OpenAI.
Developments in AI Agents
Cursor reduced agent token costs by 7% without reducing output quality by refining prompt structures, compressing file reads, improving caching, and selectively loading tools, according to @cursor_ai.
Meanwhile, Cognition integrated its Devin coding agent into Microsoft Teams and Microsoft 365. This integration links Devin directly to email, chats, files, and calendars within corporate environments, as stated by @cognition.
Vercel introduced persistent Drives in public beta for Vercel Sandbox across all account tiers. Sandboxes support up to four Drives of 16 TiB each to retain agent workspaces, code dependencies, models, and files between active executions, according to @vercel_dev.
On-Device Hardware and Chips
PrismML demonstrated its one-bit Bonsai vision-language model executing locally on smart devices powered by Snapdragon AR1 Gen 1. The setup showed fourfold lower memory consumption after optimization for the Qualcomm Hexagon NPU, as reported by @PrismML.
Finally, Liquid AI partnered with Qualcomm to run its Liquid Context system directly on the Hexagon NPU found in Snapdragon chips. This engineering step provides local personal context for autonomous agents on mobile hardware, completing today’s AI roundup overview as noted by @liquidai.
