Research
Make sense of a pile of reading
BringA question and a few articles or documents.
Ask for a short brief with a source for each key point. Separate what the sources say from what is still unknown.
A brief you can trace back to the originals.
Use AI for
everyday work
Learn how to use AI at work: turn notes into drafts, compare sources, work with spreadsheets and automate small tasks. Each workflow includes an example prompt and a way to check the result.
AI Performance Tracker
Is AI performing worse than usual?
We run the same task across different hours to track changes in AI output quality. Compare recent results with earlier runs to spot possible performance drops and look for more reliable times to use a model.
What can AI help you do?
Pick a task you already understand.
It makes a useful result easier to recognize.
Writing
Turn rough notes into a useful draft
BringYour notes, the audience and a sample of your voice.
Start with an outline. Choose the points to keep, then draft one section at a time. Ask AI to flag missing information.
An email, article or presentation outline.
Spreadsheets
Clean up a table and find the pattern
BringA copy of your table and a description of its columns.
Try a formula or cleanup on a few rows. Compare totals and check that dates, units and categories still mean the same thing.
A cleaner table, a useful formula or a chart.
Visuals
Make a picture from an idea
BringA subject, a reference and where the image will be used.
Describe the composition, colors and details that matter. Review the first result, then change one specific thing at a time.
An illustration or a visual direction to develop.
Small tools
Build something for one annoying job
BringSample input and an example of the result you want.
Ask a coding agent for a small script or page. Have it explain how to run it, then test normal, empty and unexpected inputs.
A utility for a job such as renaming files.
Everyday routines
Make the next run easier
BringThe steps of a task you repeat and one good example.
Try the process once with AI. Save the instructions, output format and checks that worked before adding automation.
A reusable method for next week’s work.
AI workflow examples
Bring your own material.
Use the example request, then check the result.
Meeting notes → a follow-up email
Draft in a chat
Start a new chat and paste the notes. Say who will read the email and keep it as a draft for review.
- Keep decisions, owners and deadlines in the notes. Remove anything you should not share with the tool.
- Ask for a short subject line and a draft that uses only those notes. Leave missing details marked as “to confirm.”
- Compare every action with the notes, then copy the reviewed draft into your email app.
An example request
Draft a follow-up email for the project team from the notes below. Include decisions, action items, owners and deadlines. Use only the notes. Mark missing details as “to confirm.” Keep it under 150 words. Do not send it. Meeting notes: [paste your notes]
Sample input
Decision: keep the November 3 launch. Maya: send copy by October 23. Homepage QA: due October 27; owner not assigned.
Example output
Subject: November 3 launch — next steps We’re keeping the November 3 launch. Maya will send copy by October 23. Homepage QA is due October 27; owner to confirm.
Before you call it done
Both deadlines match the notes. The draft leaves the QA owner unassigned.
Three sources → a comparison you can check
Start with the decision criteria
Paste short source extracts into one chat and label them A, B and C. For longer files, attach them and use their filenames as references.
- Write the decision criteria first, such as a monthly budget and a required export format.
- Ask for a comparison with a source label on each claim. Request “not stated” wherever a source is silent.
- Check the recommended option against the original sources and follow up on the missing details.
An example request
Compare sources A, B and C against these requirements: [requirements]. Cite [A], [B] or [C] for each factual claim. Write “not stated” for missing information. Recommend an option only if it meets the stated requirements, then list what I still need to check. A: [source] B: [source] C: [source]
Sample input
Need: CSV export, at most $25/month. A: $30/month; CSV export; email support. B: $20/month; CSV export; support not stated. C: $25/month; PDF export only; live chat.
Example output
B meets the price and CSV requirements [B]. A exceeds the budget [A]. C offers only PDF export [C]. Before choosing B, check its support coverage: not stated [B].
Before you call it done
B meets both stated requirements. Its support coverage remains unknown. These are sample offers, not live product prices.
A folder of files → a reusable routine
Work on a copy
Open a local test folder in the app. Keep the originals elsewhere and ask for a rename plan before approving any changes.
Try with Codex
- Set the scope to one test folder and give the naming rule. Ask for a table of current and proposed filenames.
- Review the table. Stop if a target filename already exists, a file is missing or an extension would change.
- Approve the reviewed plan on the copies. Open the files afterwards and keep the rename map so you can trace each change.
An example request
In this test folder only: [folder path], propose renaming JPG files to trip-001, trip-002 and so on, sorted by current filename. Preserve each extension exactly, including case. Leave other files alone. First show the full rename map and check whether any destination already exists. Do not change files until I approve. After approval, verify that file contents are unchanged.
Sample input
IMG_001.jpg IMG_002.JPG notes.txt
Example output
IMG_001.jpg → trip-001.jpg IMG_002.JPG → trip-002.JPG notes.txt → unchanged
Before you call it done
The two images still open and their contents are unchanged. An existing target filename must stop the rename.
AI assistants, agents and Skills
You do not need every piece
to get started.
AI assistant
You want help thinking or making a draft.
Start with a conversation. Add files or images when your task needs them, and refine the result through follow-up requests.
Ask before adding it
Can it read the material you actually use?
AI agent
You want help carrying out several steps.
An agent can work with tools, files or a browser. Define the result, give it a small scope and decide when it should stop for your review.
Ask before adding it
Can you inspect its work and control its actions?
Plugins & connections
Your information lives in another app.
A connection gives an AI tool access to a service or source. Add one when it removes a real manual step, such as finding a document.
Ask before adding it
What can it read, and what can it change?
Skills
You want to reuse a method that works.
A Skill packages task instructions and may include scripts or reference material. Support varies by tool; start with a process you have tried.
Ask before adding it
Does it fit your tool, inputs and expected output?
Try the same small task before committing to a tool. Compare the useful output, the corrections it needs, the cost and how it handles your data.
Common questions
Do I need to know how to code?
Many writing, research and document tasks can begin in a chat interface. For tasks involving code, an agent can help create and explain it. You still need to test whether the result does what you asked.
What makes a useful AI prompt?
Name the result, provide the relevant material and say what matters: the audience, format, length or constraints. Include an example when you can. If the result misses the mark, describe the specific change you need.
What is the difference between a chatbot and an agent?
A chatbot lets you work through a conversation. An agent can also take steps with tools, such as reading files or editing a document. Products often combine both. Look at what the tool can actually do and which actions you control.
How do I know whether an AI answer is right?
Check names, numbers and quotations against the originals. Open cited links. For spreadsheets, check sample rows and totals; for code, try representative inputs. A confident answer is a starting point for checking, not evidence by itself.
When should I automate a task?
When you repeat it often and can describe a correct result. Work through the process with AI first. Keep the reliable steps, decide how failures should be handled and leave judgment calls for review.