Office workflows
Choose your first AI workflow at work
Start with a frequent, low-risk task whose output you can check quickly. Test a small draft or summary, compare it with your usual process, and count review time before deciding to keep it.
An empty chat box makes AI feel like another job. A specific task gives it a purpose. Instead of asking how to “use AI at work,” choose one output you already produce: an agenda, a first draft, or a summary of non-sensitive notes.
This guide proposes a small trial. It is not a report of measured customer results.
Pick a task you can judge
Write down three tasks you repeat. For each, ask:
- Do I have the information needed to complete it?
- Can I recognize a wrong or incomplete answer?
- Can I undo or reject the output before anybody relies on it?
- Am I allowed to put this information into this tool?
A draft agenda from approved notes is a useful starting candidate. An unsupervised customer reply is a poor first experiment: errors travel further, and checking may require information you do not have.
Pick the candidate with the clearest finish line. “Help with meetings” is vague. “Turn these five topics into a 30-minute agenda, with a decision for each topic” is testable.
Give the task a small brief
Good instructions say what the output is for, what information to use, and what shape the result should take. OpenAI’s prompt engineering guide describes specifying instructions and providing relevant context. Here is an original template:
Task: Draft a 30-minute team meeting agenda.
Audience: Project team members.
Use only the notes between <notes> tags.
For each agenda item, show topic, time, and desired decision.
Keep the total at 30 minutes.
If a decision owner is missing, write “owner to confirm.”
Do not invent deadlines, people, or commitments.
<notes>
[Paste approved, non-sensitive topics here.]
</notes>Replace placeholders before using it. Add examples when they clarify what you mean. A short sample of your preferred format can be more helpful than several paragraphs of adjectives.
Compare the whole process
Try the task on a few representative examples. Record your usual completion time, time spent preparing the prompt, and time spent checking and correcting the result. These are your observations, not estimates supplied by the model.
Keep a small log with four columns: task, draft quality, corrections required, total time. Mark recurring mistakes. If every draft misses the same constraint, improve the brief once and repeat the comparison.
Do not count fast generation as a time saving when it creates slow review. Also consider whether a plain template would solve the task with less effort.
Decide: try, skip, or use
TRY means a bounded experiment is worth another look. SKIP means the task, tool, or review burden does not fit your work. USE means your own checks support using it for this specific task, with review still in place.
These labels describe a decision process, not a universal product rating. A tool can fit agenda drafting and fail at another job.
Keep the prompt, checking steps, and a good example together. Once the process is dependable, consider a second task. For a concrete next example, see turning notes into a weekly update.