Tech • AI • Robotics
ChatGPT Now Watches You... Claude Adds Watermarks (and more)
9.3/10TL;DR
A new wave of AI updates centers on more autonomous workplace agents, stronger labeling of machine-written text, expanded desktop memory features, and a faster Gemini model.
KEY POINTS
Grockbot launches as a multi-agent workspace
Grockbot is positioned as a simple interface for building and chatting with specialized AI agents on desktop and mobile. The product functions like a hub for multiple bots, including agents for health tracking, inbox management and content operations, with persistent chats and shared workflows.
Agents can complete business tasks across apps
A key feature is the ability to connect tools such as Gmail and ClickUp so agents can generate invoices, create task records, draft emails and monitor payment status. In one example, an invoicing workflow created three invoices, added them to a project board with due dates and pricing, then drafted matching emails with PDF attachments.
Virtual computers and event-based routines expand automation
Each Grockbot can use its own virtual computer, allowing it to operate software even when no direct connector exists. The platform also supports scheduled routines and trigger-based actions, such as running an inbox review every day at 8 a.m. or launching a task when a new Slack message appears in a channel.
Zapier support broadens app coverage
Native integrations are supplemented by Zapier through an MCP connection, extending access to more than 9,000 applications. That setup allows services without a built-in connector, such as Beehiiv, to be used for pulling newsletter statistics and drafting campaigns inside the same agent environment.
High pricing remains the main obstacle
The biggest limitation is cost. Access currently requires a Cursor Ultra plan at $200 per month or a Cursor Premium Teams account at $120 per month, reflecting an early beta rollout aimed at higher-tier users before broader release.
Claude begins watermarking generated text
Anthropic said Claude now adds text watermarks to generated writing to comply with the EU AI Act and the transparency code signed in July 2026. The company said the watermark does not change visible quality, add hidden characters, or increase token costs.
Watermarks are designed for attribution, not identification
Anthropic said the watermark carries no data that can identify a person, organization or specific chat. It is meant only to indicate that Claude was likely involved in producing or heavily editing a piece of text, and a detection API is expected to follow.
ChatGPT adds computer history on desktop
OpenAI introduced computer history, a desktop feature that lets ChatGPT summarize recent activity across apps and websites. The system can answer questions about recent work, reconstruct workflows and suggest automations based on what a user has been doing.
Privacy controls focus on local storage and opt-in use
The feature is off by default. Users can exclude specific apps or websites, and activity events are kept for up to 48 hours, while summary files are stored locally as Markdown until deleted. OpenAI says data is processed in a temporary server session, not stored permanently and not used for training.
Gemini 3.7 Flash arrives, but rollout is incomplete
Google has announced Gemini 3.7 Flash as an upgrade over 3.6 Flash, with benchmark gains in coding and web tasks. However, the new model was not yet broadly available in the standard Gemini chat selector at the time of the update, with early access appearing to be tied to Gemini Spark.
CONCLUSION
The latest AI releases show the market moving beyond chat into persistent agents, memory-driven assistants and content provenance tools. The main questions now are cost, privacy and how quickly these systems become reliable enough for everyday work.
Explain this