Tech • AI • Robotics
AI News: Big ChatGPT & Grok Bot Updates, Meta Muse Is Here
8/10TL;DR
Meta launched the consumer AI agent Muse, ChatGPT added writing-style personalization from connected apps, and Grokbot rolled out workflow upgrades aimed at making AI agents more practical for everyday business tasks.
KEY POINTS
Meta launches Muse
Muse, released by Meta on September 8, is a consumer AI agent designed to handle tasks across connected apps and a secure virtual browser. It can carry out browser-based actions such as booking flights and managing online workflows, placing it in the same emerging category as general-purpose AI agents for nontechnical users.
One agent, many threads
Unlike multi-agent setups that separate work across several bots, Muse centers on a single main agent with a side chat system to keep tasks organized. That design allows one identity to manage different conversations without mixing contexts, while still giving users separate threads for discrete jobs.
Feed, goals and task tracking
Muse includes a customizable feed that acts as a running brief of activity across connected applications and selected topics such as business operations or AI news. It also offers a goals feature that can create structured trackers, including health routines like daily protein intake, workouts and running targets, plus a daily task log showing what the agent and its sub-agents completed.
Heartbeat automation and approvals
Recurring jobs in Muse are organized as heartbeats, which let users schedule tasks at intervals as frequent as every 30 minutes. A dedicated approvals tab surfaces actions that require permission, such as signing into services, giving users a tighter control layer over agent autonomy.
Custom memory and social connectors
Muse can be customized through settings labeled soul and memory, which define interaction preferences and store persistent facts in markdown-style files. Its connector list includes standard tools such as Gmail, Google Drive and Calendly, but also Instagram, Instagram Messages and Facebook Messenger, opening potential use cases for social media operations and marketing.
ChatGPT learns a user’s writing style
ChatGPT introduced a new personalization feature that learns tone and phrasing from connected sources including Slack, documents and email. Users can enable reference my writing style in personalization settings, then let the system analyze prior messages and drafts to generate replies that more closely match their real communication style.
Email drafting becomes more hands-off
After analyzing connected writing samples, ChatGPT can draft replies in roughly 30 seconds and lets users edit selected text inline before sending. It can also send messages directly without forcing users back into Gmail, reducing friction for routine email work.
MCP expands app connectivity
For apps not natively supported, users can extend agent access through MCP servers, including Zapier MCP, which connects to more than 9,000 applications. The setup uses developer mode, a plugin URL and app-specific tool permissions, allowing systems such as Drip email marketing software to be controlled from within the AI workflow.
Grokbot adds three workflow upgrades
Grokbot introduced three notable updates: in-chat login completion, native email draft review and one-click switching between multiple accounts. The login flow now lets users authenticate to websites from within the chat interface, while draft review allows edits, subject changes, CC additions and direct sending without opening Gmail.
Multiple workspaces and content automation
The new account-switching feature helps separate personal and business workflows with different email-connected workspaces. In one demonstrated use case, Grokbot generated Instagram carousel content in about 2 minutes, then connected through an MCP integration to a social scheduling tool to publish posts and trigger comment-based DM automation, a process typically handled manually through tools like ManyChat.
CONCLUSION
The latest releases show AI companies racing to turn chatbots into full consumer agents that can browse, write, schedule and execute across software ecosystems. The key competition is shifting from raw model quality to how smoothly these tools handle real work with minimal user intervention.
Explain this