Tech • AI • Robotics
AI News: Big Claude & ChatGPT Updates, Grok Bot Jarvis Leak & More
7/10TL;DR
Anthropic and OpenAI expanded workplace AI features with a built-in Claude Co-work browser, shared memory across Claude modes, multi-account Gmail support, and trigger-based ChatGPT Work automations, while Grockbot appears to be testing voice calls.
KEY POINTS
Claude Co-work adds its own browser
Anthropic has added a built-in browser to Claude Co-work in the desktop app. When a task requires the web, a side-panel browser opens and Claude can navigate pages, read content, click and type without taking over the user’s main browser. The feature reduces reliance on the older browser-extension workflow and lets users hand off web tasks from within the app.
Desktop browser can be driven remotely
The browser is tied to the desktop app, but it can still be controlled from the web or a phone as long as the desktop client remains open and online. That creates a hybrid setup in which a mobile request can trigger browser actions on a connected computer. For users who work across devices, it effectively turns the desktop app into a remote execution point.
Signed-in sessions can be imported selectively
Users can import cookies to keep sites signed in inside the built-in browser. Access can be granted selectively by site, allowing some services to be included while excluding others. That approach gives Claude more practical access to authenticated workflows, while also raising the importance of careful permission management for sensitive accounts.
Shared memory now spans Claude chat and Co-work
Anthropic has merged memory across Claude chat and Claude Co-work, ending the split between the two modes. Saved preferences and context can now carry over automatically, reducing repeated setup and re-explanation. Users can also review, edit or delete stored memories directly in settings.
Memory is organized and user-editable
Saved memory can include categories such as preferences, finances, fitness, photography and other recurring topics. A preference edited in memory is then available in both chat and co-work interactions, showing that the system is now synchronized across interfaces. The update is especially significant for users who move between quick conversations and longer task-based sessions.
Zapier MCP broadens app connectivity
AI agents remain limited by native integrations, but Zapier MCP offers a workaround by linking them to more than 9,000 apps. After creating an MCP server and adding tools for selected services, users can connect that setup to agents such as Grockbot. The result is a much broader operational surface for tasks involving platforms not directly supported by default.
ChatGPT now supports multiple Gmail accounts
ChatGPT can now connect multiple Gmail accounts under one plugin, allowing users to manage personal and business inboxes in the same environment. After connecting each account and assigning labels or nicknames, users can read, sort, draft and send email across inboxes. The feature is likely to appeal to operators juggling several addresses each day.
ChatGPT Work introduces trigger-based automations
ChatGPT Work now supports tasks that run not only on schedules but also when specific events occur in connected apps. A demonstrated workflow used a Gmail trigger to detect a likely sponsorship inquiry and automatically post a concise summary into a designated Slack channel. The generated alert included sender details, subject, opportunity context and a reference to the email thread.
Current automation scope is narrow but useful
Triggered automations currently work with Gmail, Slack and GitHub. Example use cases include drafting replies when client emails arrive, summarizing active Slack threads, reviewing code when a pull request opens, or sending daily briefings at set times. The feature is currently available on the web and mobile apps, but not yet in the desktop app.
Grockbot voice mode appears to be in testing
A leaked interface image suggests Grockbot may be getting a voice-call mode, indicated by a headphones button for calling custom agents. No official release has been confirmed, but the addition would move the product closer to hands-free agent management. If launched, voice interaction could become a major shift for users running multiple agents in parallel.
CONCLUSION
The latest updates show enterprise AI tools moving beyond chat into persistent memory, direct web action and app-driven automation. The competitive focus is increasingly on making assistants act across software systems with less manual coordination.
Explain this