Video summary
ChatGPT Work Is More Powerful Than You Think
Main summary
Key takeaways
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).