Blog · Writing with AI
What actually saves time when you write with AI on a Mac
The slow part of AI-assisted writing is rarely the AI. It is everything around it. Here is where the time goes, and what helps.
ToneMate ·
In shortWriting with AI saves time when it stays out of your way: the text is handled where you already are, you pick the smallest action that does the job, you give the model the one piece of context that matters, and you read the result before it goes out. We have no measurements to quote, so at the end there is a ten-message test you can run on your own work.
Where the time goes
Think about the last time you asked an AI to improve a message. You probably left the app you were writing in, opened another tool, pasted your text, explained who it was for, waited, read the answer, copied it, went back and pasted it again. The model's part was a few seconds. The rest was switching and re-explaining.
That loop is where most of the time goes, and it is why the same model can feel fast in one setup and slow in another.
Keep the text where it is
The first habit is to act on the text in place: select it, run an action, insert the result. ToneMate is built around that flow, with one shortcut per action. To be accurate about how it works, it copies your selection with the clipboard and pastes the result back, and it restores what you had copied before, so it works where an app lets you select, copy and paste text.
Whatever tool you use, the test is the same: count the steps between “I have a draft” and “the better draft is in place”.
Pick the smallest action that does the job
A full rewrite is slower to review than a small fix. Match the action to the problem:
- Typos and grammar: proofread, and read the changes.
- A message that sounds wrong: change the tone, and keep the facts.
- Too long or too short: expand or shorten, then check what moved.
- A long text you need to digest: summarize it, and keep the original at hand.
- A blank page: generate a first draft, then edit it.
Say what you mean, then let it help
When you ask for a reply, a sentence about what you want to say changes the draft more than any tone setting. When you have a lot to write, saying it aloud and cleaning up the transcript is often easier than starting from the keyboard. Both are about giving the model the one thing it cannot guess.
Decide where your text goes
Some text you would rather not send to a cloud provider. A model running on your Mac keeps it there, at the price of some setup and a smaller model. A cloud model is more capable and bills your own API usage. Being able to choose per task, rather than once for everything, saves the argument with yourself.
Read before you send
Models slip: a changed date, an invented detail, a tone that is slightly off. Time saved on drafting is lost if you have to apologise for a mistake, so build a half-minute read into the loop. Seeing the original next to the result, with the changes marked, makes that read quick.
Test it on your own work
We would rather not hand you a number, because the right one depends on you. Try this instead.
- Pick ten real messages or paragraphs you would normally write or fix by hand.
- Time five of them the way you do today, and five with the tool.
- Note the time, and also how many times you had to correct the result.
- If the second group is not clearly better for you, change the tool or the habit, not the test.