Video summary

ChatGPT Work Is More Powerful Than You Think

Main summary

Key takeaways

Technology

Summary of technological concepts & Work features (from the subtitles)

Work vs Chat improves output quality and format

  • Using ChatGPT Work can yield better results than standard Chat for the same prompt because Work is designed to:
    • take longer
    • break work into tasks
    • use tools automatically
    • produce structured deliverables by default
  • Example (Disneyland planning request):
    • Chat: faster response (~under 5 minutes), includes images and a ranked/organized plan (good, but more “conversation-like”).
    • Work: slower response (~10+ minutes) that generated a formatted, color-coded document with checklists/tables, highlighted important items (red), and provided helpful links—document creation happened without explicitly asking.

Multitasking via concurrent Work sessions

  • Work allows multiple sessions running at once, so long-running tasks can continue while you start other work threads.

Unique Work capability: “Sites” (hosted interactive websites)

  • Work can create a live, fully hosted website (hosted by ChatGPT).
  • The site can be:
    • made public via a share URL
    • customized with:
      • Custom URL prefix (and potentially a custom domain with configuration)
      • Environment variables, including choosing which model the site uses
  • Example outcome:
    • a “Disneyland family trip command center” with editable sections such as hotel recommendations, day-by-day plan, adjustable budget, and a packing checklist—shared through a live URL.

Execution model: Work runs in a “cloud computer” per task

  • Each Work task gets its own cloud execution environment, enabling it to:
    • spend more time
    • take more actions
    • maintain progress through the session
  • Work also syncs to mobile, letting you start from a phone and continue later without keeping a browser tab open.

Plugins: extend Work with external tools

  • Plugins are enabled via a plugin panel where you can search for and add them, including handling required logins/permissions.
  • The workflow is improved versus earlier plugin versions:
    • less clunky
    • the model uses plugins more fluidly
    • plugins include “skills” to teach the model how to use the external tool properly
  • Example multi-plugin automation:
    • gather context from Google Calendar, Gmail, and meeting notes to prepare a requirements-based 8-slide presentation
    • output produced as a PowerPoint file
    • verify/edit for issues before finishing
  • The system can adapt even if the prompt is intentionally vague, with plugins supplying missing context and tool access.

Desktop app + local-file workflows

  • Some plugins work only with the desktop app, especially when they require:
    • local files
    • terminal access
    • a more capable browsing/editor environment
  • Example plugin: Remotion (editable motion graphics)
    • generate a motion graphic in Work, then refine directly in the editor (e.g., manual adjustments when elements overlap)
    • uses explicit plugin selection via @remotion to disambiguate between competing plugins (e.g., comparisons involving Hyperframes).

In-app browser (native browsing + site interaction)

  • Work includes a browser inside the app that supports:
    • human-like browsing: open sites, click, fill forms, navigate, and leave tabs open
    • a better experience than older browser approaches (e.g., improved versus a Chrome extension)
  • Example:
    • research land listings in Utah, apply filters (forested, near adventure areas, acreage range), and open results in separate tabs with results presented in a table
  • Also supports:
    • cookie/password import for logged-in browsing, enabling persistent sessions and interactions.

“Blocks”: reusable interactive UI-like outputs

  • Work can generate blocks that are editable and sometimes interactive/sendable:
    • Writing blocks: draft an email/message and edit inline
    • Email block: can send via connected Gmail
    • Code/interactive blocks, such as:
      • Kanban board
      • Dashboard
      • Calculators (ROI/compound interest examples)
      • Timelines, flowcharts, tier lists, etc.
  • Blocks reinforce the idea that Work outputs can be deliverables, not just text.

Branching chats

  • A feature to fork from any earlier message (“branch a new chat”) to explore alternate paths while keeping the original thread intact.

Skills (workflow packaging for reuse)

  • Skills are recipe-like reusable workflow instructions that teach Work how to repeat complex multistep tasks.
  • Example:
    • creating a “Remotion visual style” skill so future requests don’t require long prompts
  • Skills may be migrated from other platforms (e.g., download from Claude, then convert/drop into ChatGPT to turn into a skill).

Scheduled tasks (automation over time)

  • Work supports scheduled automation:
    • one-off or recurring schedules (daily/weekly/hourly)
    • fully automated execution once created
  • Recommended approach:
    • create a skill first, then schedule it for consistent results
  • Practical guidance:
    • prefer scheduling in cloud so your computer doesn’t need to stay on
    • if local files are needed, use Google Drive + plugins so Work can access them when scheduled

Main speakers / sources

  • Speaker: The video is presented by a single narrator/host (first-person demonstration).
  • Sources mentioned in the content:
    • ChatGPT Work
    • ChatGPT Sites
    • Codex
    • plugins (including Remotion and a comparison to Hyperframes)
    • external services like Google Calendar, Gmail, and Granola (as a referenced notes tool).

Original video