Musk’s Grok Bot has released version 1.0. This time the idea is to design the robot as a colleague.
The biggest change is sharing. A team can share the same robot, chat with it in Slack, Microsoft Teams, or drag it into a video conference, where it participates in discussions through the chat box in the conference.
The second change is initiative. It watches for signs at work: a deadline is approaching, a task is half-finished, new research doesn’t match the team’s original plan. After discovering it, it will take the initiative to ask whether it wants to take over; when it encounters a place that needs someone to make a decision, it will bring in a real person.
Another design is quite clever: someone on the team teaches it a better way of doing things, and everyone who uses it benefits. Experience is no longer locked into one person's conversation log.
The authority also raised the issue of permissions. The robot will comply with each person's respective authority scope and will ask questions before performing sensitive operations. However, there is a natural tension between "taking the initiative to find work" and "not overstepping one's authority": how it handles tasks that it shouldn't touch can only be known after it has been used by a real team for a period of time.
The one-month-old version of Grok Bot has been iterating very quickly, from having its own mailbox, to being open to more packages, and now to the team-shared version. In the same week, Claude also launched the function of entering meetings as a participant. The competition among office agents is shifting from "whose model is stronger" to "who is more like a reliable colleague."