Most advice about using AI at work is a list of prompts. Prompts are the easy part. What separates people who get real value from people who lose an afternoon is that they run the same loop every time.
The loop
1. Decide what good looks like. Before you type anything, know how you would recognise a correct answer. "A summary" is not a target. "Four bullets a colleague could act on without reading the original" is. If you cannot describe the target, you cannot judge the output, and you will accept whatever sounds confident.
2. Give it the material. Most disappointing results come from asking a model to work from memory when you could have handed it the actual document, code, or data. Paste the source. This single habit removes a large share of invented detail.
3. Work in passes. Ask for the structure first, correct it, then ask for the content. One large request produces one large thing to fix. Three small ones let you catch a wrong turn on step one instead of discovering it at the end.
4. Verify what matters. Decide which claims you would be embarrassed to be wrong about, and check those. Not everything — that would cost more than doing the work yourself — but every number, name, date and citation that will leave your desk.
Where AI actually saves time
Be honest about the shape of the saving. It is largest when:
- There is a blank page and starting is the hard part
- The work is transformation — same information, different form
- You need volume for selection — ten options so you can pick one
- You are outside your specialism and need orientation before you know the questions
It is smallest, and often negative, when the task needs facts you must then check one by one, or when you already know exactly what you want and typing it is the only work.
Knowing when to stop
Set a limit before you start — three attempts, ten minutes. If you pass it, the task is a poor fit, your target was unclear, or the model lacks something it needs. Continuing to reword past that point is the most common way people lose time while feeling productive.
The compounding part
Keep what works. A prompt that produced a good result once will produce it again next week, and you will not remember the wording. People who save their working prompts get faster every month. People who start from a blank chat every time stay exactly as fast as they were on day one.