With OpenAI’s current wave of Work and Codex improvements, I wanted to contribute a collection of productivity and efficiency suggestions gathered through extensive use of ChatGPT, Desktop, Work, Codex, Apps, Skills, and MCP integrations.
Many of these suggestions address relatively small frustrations that accumulate into substantial losses of time, computing resources, and productivity.
The underlying principle is simple:
Users shouldn’t have to repeatedly reconstruct work, resources, or workflows that already exist within the ChatGPT ecosystem.
1. Skills, MCP, and App Integration
1. Universal Skill Library
Allow Skills to be shared across compatible ChatGPT environments, including Desktop, Work, Codex, and Apps, without repeatedly uploading or installing them.
2. Skill Bundle → App Conversion
Introduce a way to turn existing Skill bundles into reusable Apps, or wrap them with an App interface. Existing instructions, references, and supported scripts could be reused rather than reconstructed.
3. Local MCP for Personal Apps and Websites
Enable secure MCP connections between ChatGPT and user-created websites or locally hosted applications. This would allow ChatGPT to interact with applications directly, subject to explicit user permissions.
4. Shared MCP Connections
Allow compatible Apps and agents to reuse existing MCP configurations without each requiring a separate integration setup.
5. Local Side-Panel Apps
Introduce small, dockable utility Apps similar to browser side panels. Examples include calculators, prompt libraries, note organizers, and reference tools that remain accessible while working in ChatGPT Desktop.
6. Local ↔ Cloud Workspace Transfer
Provide a convenient way to move or branch local Desktop projects, pages, and resources into cloud workspaces, and vice versa, without manually rebuilding or re-uploading their contents.
2. Library and Information Retrieval
7. Library ↔ Project Folder Linking
Allow Project folders to reference existing Library folders. This would make associated images, documents, Skills, and other resources immediately accessible without creating duplicate copies.
8. Improved Add from Library Interface
The Library picker needs better organizational and filtering capabilities, particularly for users with large collections.
Suggested improvements:
-
Browse by folder.
-
Filter by file type, especially images versus documents.
-
Thumbnail grids and previews for images.
-
Sort by name, date, and recent usage.
-
Search within specific folders.
Currently, important resources can become buried among unrelated files, sometimes making manual re-uploading easier than finding the original.
9. Advanced ChatGPT Search
Expand conversation and file searching with additional parameters, including date ranges, exclusions, project location, file type, and exact phrases.
Search results should also make it easier to identify which conversation, file, or source location contains the matching information.
10. Intelligent Conversation Renaming
Conversation titles often reflect the opening question rather than the subject the conversation eventually develops into.
Allow ChatGPT to suggest updated titles as the central topic emerges, ideally with an option to approve or disable automatic renaming.
3. Computing Efficiency and Usage Management
These are particularly important for users operating under limited weekly usage allowances.
11. Automatic Thinking-Effort Downshifting
Users sometimes leave their thinking settings on High, then use expensive reasoning capacity for relatively simple questions.
An optional adaptive mode could recognize when a task requires less effort and temporarily downshift, while preserving manual overrides.
12. Astra / High-Compute Usage Gauge
Before beginning a potentially expensive task, provide an estimated usage range or warning about how much of the remaining allowance the task might consume.
Even approximate estimates would help users decide whether to commit significant resources.
For example:
“This task may consume a substantial portion of your remaining weekly allowance. Continue, reduce scope, or use a lower-compute model?”
The estimate need not be perfectly accurate. Clearly labeled uncertainty would still be preferable to discovering the cost afterward.
13. Optional Verification Controls
Some high-compute workflows can spend significant resources on repeated verification, even when the user primarily wants exploration, ideation, or preliminary analysis.
Introduce controls for optional verification passes, priorities, and compute budgets.
Essential correctness and safety checks should remain in place, while additional discretionary verification could be user-configurable.
14. Usage and Failure Analytics
Provide clearer information about which tasks, agent operations, verification passes, and retries consume significant portions of available usage.
This would help users understand inefficient workflows and make better decisions about resource allocation.
4. Desktop Productivity and Interactive Workspaces
15. Improved Floating and Side Chats
Allow Desktop conversations to be freely positioned, resized, minimized, and docked.
Floating chats should not obstruct primary conversation controls. Supporting multiple simultaneous workspaces would improve multitasking.
16. Interactive Decision Tables and Mini-Canvases
Introduce lightweight interactive tables, relationship diagrams, and planning surfaces within conversations.
Users could rearrange rows, reorder priorities, move diagram elements, compare alternatives, and let ChatGPT analyze those changes.
Think of a miniature Miro-style workspace designed for quick collaborative reasoning rather than extensive diagram production.
17. Native Prompt Library
Provide a searchable, reusable prompt manager with templates, categories, editable instruction blocks, and quick insertion into conversations.
This would reduce dependence on third-party browser extensions and improve consistency across Desktop and web.
18. Improved Desktop Accessibility
Expand typography and layout customization, including additional font choices, improved readability controls, and better management of floating interface elements.
5. Workflow Reliability and Project Continuity
19. Persistent Project Handoffs
Allow compatible agents and environments to share project context, decisions, instructions, and relevant resources without requiring users to repeatedly explain the same architecture.
20. App Editing and Preservation Safeguards
When an AI modifies an existing application, it should prioritize preserving established functionality and design requirements.
Useful safeguards could include change previews, checkpoints, rollback, and explicit confirmation before major structural replacements.
Repeatedly regenerating an unwanted design can consume substantial usage.
21. Permission and Connection Diagnostics
Clearly show which resources an App or agent can access, what permissions are missing, and what action is necessary to resolve a connection problem.
Users should not have to repeatedly troubleshoot whether an issue comes from permissions, configuration, unsupported capabilities, or missing resources.
22. Better Tool Availability Across ChatGPT Environments
Where technically supported and permitted, allow connected tools to be used across ordinary chats, Work, Desktop, and related environments without unnecessary workflow switching.
6. A Broader Architectural Opportunity
Many of these proposals could be addressed through a shared resource and workflow layer connecting:
-
ChatGPT and Desktop
-
Work and Codex
-
Skills and Apps
-
MCP integrations
-
Library assets and Projects
-
Reusable prompts and project instructions
The goal would not be unrestricted sharing. Permissions, security boundaries, and user control would remain essential.
Instead, the objective is interoperability without unnecessary duplication.
A user should be able to create a resource once, authorize its use in another environment, and continue working without reconstructing it.
7. The Overall Efficiency Principle
These suggestions come from recurring practical problems rather than an attempt to redesign the entire platform.
The desired improvements can be summarized into four principles:
-
Reuse completed work instead of reconstructing it.
-
Reference existing resources instead of duplicating them.
-
Match computing effort to the actual requirements of a task.
-
Give users more direct control over their working environments.
Some improvements are relatively small interface changes. Others require deeper architectural work.
Even implementing a portion of these would reduce friction, conserve usage allowances, and improve productivity for users building applications, researching complex subjects, or managing large creative and technical projects.
Thank you to the OpenAI teams for the continuing improvements to ChatGPT, Work, and Codex. I hope some of these suggestions prove useful during the current development push.