Integrate Codex into Work—Don’t Restrict Chat
A Product Proposal for Two Modes and Separate Usage Allowances
To the OpenAI Product Team
Core Recommendation
I recommend simplifying the product experience around two modes: Chat for thinking and discussion, and Work for getting things done—with a separate usage allowance for each.
What should be consolidated is not the usage of Chat and Work, but Codex as a separate product entry point and a separate allowance system. Codex’s coding capabilities can be fully incorporated into Work as one of its specialized execution capabilities.
Programming is work. If Work is designed for task execution, there is no need to treat coding as a separate, parallel product with its own allowance system.
Unify the product experience, not every usage budget.
1. Two Modes, Two Separate Allowances
Chat: Protect Thinking and Discussion
Chat is where users reason, research, analyze, learn, discuss ideas, and work through problems over multiple rounds. Its value is not limited to answering individual questions; it gives people room to think with AI, explore possibilities, and reach better-informed conclusions.
Chat should have its own generous and predictable allowance for advanced models. Usage in Work should not consume Chat capacity or make users hesitate to discuss a problem because they are trying to save resources for a later task.
Reasonable usage limits are acceptable, but they should be transparent and predictable—not imposed through silent model downgrades.
Work: One Place for Every Execution Task
Work should cover any task that users ask AI to carry out, including:
-
Programming, code changes, debugging, and testing;
-
File and project work;
-
Browser and computer operation;
-
Multi-step agent tasks;
-
Building, deployment, and other tool-driven execution.
Codex can be fully integrated into Work as its coding engine or specialized coding capability, rather than remaining a separate product alongside Work.
Work should have its own allowance, metered according to the model and execution resources consumed. The distinction should be simple for users: use Chat to think and discuss; use Work to have AI do the work.
2. Why a Shared Allowance Should Not Come at the Expense of Chat
1. Thinking and execution are different kinds of usage
A user may spend hours discussing a complex problem—asking questions, comparing approaches, developing an argument, and revising conclusions. That sustained exchange is itself a core part of ChatGPT’s value.
Work, by contrast, is where users delegate concrete tasks to AI: writing code, handling files, operating tools, and completing multi-step workflows.
If both modes draw from one shared allowance, execution tasks can consume the capacity users would otherwise use for thinking and discussion. Users then have to budget every conversation, making sustained exploration feel costly and constrained.
2. Put cost controls in Work, not in restrictions on Chat
Work is the more appropriate place for detailed usage metering. OpenAI can account for model choice, task size, tool calls, and execution time when designing Work allowances and pricing.
But controlling execution costs should not mean restricting advanced reasoning and sustained conversation in Chat. A user’s willingness to pay for AI to execute work does not mean they are willing to give up the space to think.
3. The product boundaries should reflect how people actually work
Users should not have to decide whether a task belongs to ChatGPT, Codex, or Work, or manage complicated allowance relationships between them.
A more natural structure is:
-
Chat: discuss, research, reason, and make decisions with AI;
-
Work: delegate tasks to AI—including programming.
This simplifies the product while allowing Codex capabilities to become part of a broader agent-based work environment.
3. Moderate Price Adjustments Can Support a Clearer, More Sustainable Experience
OpenAI does not have to preserve today’s pricing indefinitely. A reasonable price increase and clearer plan tiers are preferable to merging allowances in a way that degrades the Chat experience.
The following is an illustrative product proposal, not official pricing:
| Plan | Illustrative monthly price | Chat allowance | Work allowance |
|---|---|---|---|
| Plus | $25–30 | Generous | Basic |
| Pro | $100 | More generous | High |
| Ultra | $200 | Very high | Very high |
Each plan should provide distinct Chat and Work allowances. The amounts, reset periods, and usage rules should be clearly displayed so users can understand and plan their usage.
4. Recommended Product Principles
-
Keep Chat and Work as the two user-facing modes. Chat protects sustained thinking and discussion; Work handles task execution.
-
Remove Codex as a separate product and allowance boundary. Integrate its coding capabilities fully into Work without removing or weakening the underlying technology.
-
Maintain separate Chat and Work allowances. Meter and display them independently; neither should consume the other’s capacity.
-
Keep Chat generous, stable, and transparent. Work usage should not reduce access to advanced reasoning in Chat.
-
Meter Work according to execution resources. Use clear allowances and pricing tiers to cover its costs.
Conclusion
ChatGPT’s value is not only that it can complete tasks for users. It is also a partner for ongoing thinking, research, learning, and discussion. Chat is foundational to that relationship; it should not become whatever capacity remains after Work has consumed its share.
OpenAI should simplify the product structure by integrating Codex into Work—not combining Chat and Work into one allowance, and certainly not restricting Chat to make room for execution usage.
Programming is work, and Work can fully encompass Codex. The value of thinking and discussion in Chat, however, should be protected independently.
A two-mode product with two separate allowances, supported by reasonable pricing, would better balance user experience, resource efficiency, and long-term commercial sustainability.