Matthew Berman presents Grokbot as a conversational agent system designed for a broad audience. Each thread represents a persistent agent, but the interface hides model selection, code, file structures and tool-call details. Users can create focused bots for email, calendars, research, shopping or operations through a guided chat rather than configuring a developer environment.
Every bot receives its own cloud computer and can browse sites or work with files while sharing authentication across environments. A user can open the remote desktop, take over manually and then return control to the agent. Grokbot can also inspect a local computer, creating a hybrid workflow across cloud and local systems.
A chief-of-staff pattern lets one primary agent delegate work to specialist agents and preserves the conversations between them. In the demonstration, an email agent performs a dry-run inbox scan and reports back without archiving anything, illustrating how a coordinating agent can gather results while keeping consequential actions under user control.
Plugins connect services such as Gmail, Drive, Calendar, Slack, Notion and Box. Routines schedule repeated natural-language tasks, while a teach-a-task feature observes a manual workflow and converts it into a reusable skill. Berman notes that separating Grokbot from Cursor creates ambiguity about where some work belongs, but argues that the simplified presentation points toward agents that choose the appropriate interface automatically. The web-publishing sponsorship is omitted.
Watch on YouTube
