Choose Chat, Work, or Codex for the kind of work you need, then select an available model and judge the finished result.
Separate the workspace from the model
The names describe different choices. Chat suits questions, discussion, and quick drafts. ChatGPT Work is an experience for carrying a task through to a reviewable deliverable. Codex provides developer-oriented tools and views for software work. GPT-6 Astra is a model that can power work, rather than another task-management interface.
Available experiences, models, tools, and settings depend on your plan, client, rollout, and workspace permissions. Check what your own account offers. You do not need every option to learn the central skill: describe a useful outcome, provide evidence, and inspect what comes back.
Start with a task you can personally assess
For a beginner, use Chat to turn a messy idea into a clear brief. Suppose you are preparing a small educational workshop. Supply the audience, confirmed format, topics, date if decided, and what attendees should leave with. Ask for missing decisions before asking for promotional copy.
Then move to Work when you want a finished planning artifact. Provide the approved brief and ask for an editable launch checklist with owners, dependencies, and a separate list of unresolved choices. Work can use files and approved tools available in the conversation. It still needs access to the source material; a model cannot infer what is inside an unconnected account.
Give the task a practical definition of done
“Plan my launch” leaves too many choices unstated. Specify the audience, deliverable format, available time, confirmed facts, and what requires your judgment. Define the review: every date must match the brief, each action needs an owner or an explicit owner-needed label, and nothing may claim registrations or customer results you have not supplied.
If you already use AI daily, add an example of an acceptable output and one failure to avoid. A previous checklist reveals much more about your preferences than “make it professional.” Keep examples short enough that their useful structure remains obvious.
Use Codex when your next step involves software
Choose a small project with a visible behavior: a local page that sorts fictional enquiries by follow-up date. Supply the folder or repository, explain its existing structure if known, and ask Codex to inspect it before proposing changes. Require a demonstration of sorting and an honest explanation of whether data survives refresh.
A first project should teach you how to review changes. Open the result, try an empty list and two enquiries with the same date, then inspect the changed files. Ask what was tested and what remains unverified. Publishing, storing customer records, and connecting email are separate decisions that need a defined scope.
Choose Astra by the problem, then verify the outcome
OpenAI positions Astra for demanding work across research, apps, and code. Its current guidance also describes Sol and Luna options for different workloads. Begin with the available default unless you have a specific reason to change it. Compare models using the same brief and acceptance criteria, not a single impressive sentence.
For your launch checklist, open the actual file. Spot-check three dates, follow a dependency from start to finish, and find every unresolved decision. Record corrections and time spent reviewing. If the output misses essential context, supply that context first. A stronger model or higher reasoning setting cannot replace an absent policy, a missing file, or a decision only you can make.
Prompts to put it into practice
Replace the brackets with your context. Use information you have permission to share.
Create a reviewable launch checklist
Use my attached approved workshop brief to create an editable launch checklist. Include task, owner, due date, dependency, and completion evidence. Use only confirmed dates and promises from the brief. Put undecided items in a separate decisions list rather than choosing for me. I have three hours this week, so identify the smallest workable first sequence. Check the finished file against the brief, flag conflicts, and give me the file plus a short review note. Do not publish or send anything.
Try it on your work
Write a one-page brief for a real task. Create one deliverable, inspect the artifact itself, and record three corrections that would improve the next run.
Check the result
- You can explain which experience you chose and why.
- Every factual promise in the deliverable has supplied support.
- The output opens and is usable in its intended format.
- Missing decisions and unverified work are visible.
Sources & further reading
Product documentation checked October 16, 2026. Features, labels, and access can change. Examples and practice prompts are original AI Teachers teaching material.