Real-world examples
- Draft a follow-up email to a department committee
- Summarize meeting notes before a faculty discussion
- Answer a student's basic question about syllabus formatting
ChatGPT
Modes
Practical guidance for choosing the right mode for common campus work.
Real-world examples
- Compare two course redesign approaches and recommend one
- Outline a grant proposal section with supporting rationale
- Identify common themes across a set of student feedback responses
Real-world examples
- Analyze a budget reallocation across competing departmental priorities
- Evaluate a policy proposal with multiple stakeholder interests
- Design a research study with methodological tradeoffs
Real-world examples
- Draft, review, and refine a department communication plan in one session
- Generate a structured literature synthesis with citations from uploaded sources
- Build a multi-section report from a collection of uploaded documents
Real-world examples
- Write a research funding strategy memo with supporting evidence
- Synthesize competing findings across multiple studies for a policy brief
- Review and critique a draft policy document for gaps and inconsistencies
Work & Codex
Work & Codex Usage and Credits
Work and Codex share your monthly credit allocation and use token-based credit pricing. Usage can increase with higher-capability models, larger tasks, reasoning effort, speed settings, and other agent activity. Check your remaining monthly allowance before starting larger Work or Codex tasks.
Codex
Session Calibration
Choose your starting settings intentionally. An unsuitable default can increase credit usage more than a single complex task.
Everyday setup
- Model: Lower-cost coding model
- Speed: Standard
- Thinking: Low
Everyday anti-pattern
- Model: GPT-5.5 / latest
- Speed: Fast
- Thinking: High or XHigh
Note: Don't make the highest-capability model and Fast mode your everyday default. Fast prioritizes faster execution and may consume credits at a higher rate than Standard.
Use it when faster turnaround is important for the task. See the OpenAI speed guidance for current model support and details.
Model Guidance
Choose a current supported model that matches the task. Use a lower-cost model for routine coding and a higher-capability model when the work requires stronger reasoning or more complex agentic workflows.
| Situation | Recommended model | Why |
|---|---|---|
| Normal coding | Lower-cost coding model | Strong performance without selecting a higher-capability model unnecessarily |
| Small edits | Lower-cost coding model | Good enough for quick changes |
| Debugging across files | Higher-capability model | Needs stronger reasoning |
| Architecture, migrations, hard bugs | Higher-capability model | Appropriate when the task requires stronger capabilities |
| Repeated failures | Upgrade model | One strong attempt can be cheaper than many weak ones |
Speed Guidance
Fast mode prioritizes faster response and consumes credits at a higher rate than Standard.
| Speed | Use when | Avoid when |
|---|---|---|
| Standard | Default for almost all coding | Don't avoid it just because Fast exists |
| Fast | Live pairing, demos, urgent fixes, short tasks where waiting is the problem | Normal development, long repo analysis, large refactors, exploratory debugging |
Thinking Level Guidance
| Level | Use for | Token risk |
|---|---|---|
| Low | Rename, formatting, tiny local changes | May miss context |
| Medium | Default coding, tests, normal bug fixes, small features | Best balance |
| High | Hard bugs, multi-file reasoning, security-sensitive changes | Use when deeper reasoning benefits the task |
| XHigh | Big migrations, architecture, autonomous repo work | Use when the task requires the deepest reasoning |
Gotchas
Things to watch
Keep in mind
- The newest model is not always the best default for cost control.
- Fast mode increases speed, not coding quality.
- Use higher thinking levels when the task benefits from deeper reasoning.
- Many cheap retries can cost more than one well-scoped, higher-effort request.
- Open files and selected code matter. Codex uses VS Code context, so keep relevant files open to write shorter prompts and get better results.
Best Prompt Patterns
Default setup
Everyday codingCopied
Use the open files only. Make the smallest
safe change. Do not refactor unrelated code.
Explain what changed and list any tests
I should run.
Harder work
Complex tasksCopied
Analyze the relevant files first. Propose
a short plan before editing.
Make the smallest safe change.
Preserve existing behavior unless required.Bottom Line
| Task | Model | Speed | Thinking |
|---|---|---|---|
| Everyday coding | Stable coding model | Standard | Medium / Low |
| Harder coding | Stable or latest model | Standard | High |
| Urgent coding | Any | Fast (briefly, then switch back) | As needed |
