Cursor just made something incredible...

Summary of Technological Concepts & Product Features (Grockbot)

• Grockbot is a new agentic AI app designed to feel like a natural chat conversation with an AI agent.
• It targets a mass audience and is not positioned as a developer-focused coding tool.

Native Chat-Style UI

• The UI presents a chat experience similar to Telegram/WhatsApp:
  • Threads appear on the left.
  • The conversation appears on the right.
• Key contrast with coding/agent tools (e.g., Cursor/Codecs/Claw):
  • The app does not show code, directory structures, or visible tool calls.
  • The “thinking” experience is simplified as conversational state (no shimmers or explicit execution traces).

Key Behavioral / Architecture Details

Thread Model: Thread = Individual Agent Concept

• The app uses threads where each thread corresponds to an agent.
• However, the reviewer’s expectation is closer to threading by topic (a more Cursor-like mental model) rather than “each thread is an agent.”
• Even if both models are similar behind the scenes, the review highlights this mismatch.

“Chief of Staff” Workflow (Reviewer Adaptation)

• The reviewer adapts the agent-per-thread model into a single continuous agent experience using:
  • A main agent acting as “chief of staff” to manage and coordinate other capabilities.

Live Environment: Cloud + Local Hybrid Control

Isolated Per-Bot Compute Environment

• Bots can spin up a full isolated computing environment, which appears like a Linux desktop with a GUI file manager.

Shared Authentication, Separate Environments

• Each bot has a separate environment, but authentication can be shared.
• Example: a user can browse Amazon without re-login for each bot/environment.

Local Machine Control

• Grockbot can also query local machine information (e.g., counting desktop folders), creating a hybrid local + cloud agent model.

Multi-Agent Communication & Persistent Interactions

• Agents can message each other depending on intent.
  • Example: when discussing inbox handling, Grockbot may route work to an email agent for scanning/triage.
• Agent-to-agent conversation persists indefinitely, preserving accumulated context.
  • The reviewer demonstrates viewing an embedded conversation between the chief-of-staff agent and the email agent, showing observable internal coordination.

UX / Capability Choices for Non-Technical Users

• No model picker: users cannot choose among model variants (e.g., GPT versions).
• No visible tool calling / no code artifacts:
  • Explicitly contrasted with Cursor’s more transparent “explored files / terminal / tool calls” style.
• Potential downside raised by the reviewer:
  • The boundary between Grockbot (knowledge work) and Cursor (coding) could become blurred, prompting the question: which app should handle what—unlike “super-app” approaches (noted via an example referencing OpenAI tool merging).

Extensibility: Plugins / Integrations

• A plugins marketplace is a prominent feature (conceptually similar to MCP servers).
• Mentioned plugin examples:
  • Gmail
  • Google Drive
  • Google Calendar
  • Slack
  • Notion
  • Box
• Typical integration flow:
  • Users often authenticate once, after which agents can use the integration for actions.

Automation Features: Routines and Tasks

Routines (Recurring Scheduled Behaviors)

• Users create a routine per agent via a UI flow involving agent profile → routines.
• A routine includes:
  • Natural language instruction
  • Schedule (e.g., every Monday at 8:00 AM)
  • Active/inactive toggle
  • Test run
• Example routine behaviors:
  • Weekly “computer cleanup” suggestions of stale files
  • Daily/recurring email summaries for urgent replies

Tasks (Teach the Agent by Demonstration)

• A “Teach a task” flow:
  • The agent operates inside its own environment/computer.
  • The user demonstrates steps (e.g., open a camera product URL and capture price into a sheet).
  • Grockbot generates a reusable skill from the demonstration.
• Example:
  • Create a skill to log Amazon camera prices into a Google Sheet and update the sheet based on observed pricing.

Skills (Reusable Capabilities) + Example “Humanizer”

• Grockbot can create skills from demos (“Teach a task”) and also supports installing skills externally.
• Example mentioned:
  • “Humanizer” skill to make AI writing less “AI-like” (described as “AI slop”).

Main Speakers / Sources (as implied by subtitles)

• Matthew (reviewer; appears addressing the audience)
• Grockbot team / Cursor team (referenced in connection with the chief-of-staff agent idea)
• Here.com (video sponsor for publishing documents/public links)
• Cursor / Codecs / Claw (comparison products referenced)
• Email provider: Gmail
• Cloud apps/integrations: Slack, Notion, Box, Google Drive/Calendar (listed as plugin examples)