The reply I avoided for three days took three minutes

The message is not long. That is almost what makes it worse.

A client wants to know why the work is late. A recruiter asks for a follow-up. A relative sends something kind that deserves more than a thumbs-up. You read it, decide you need the right words and leave it for later.

The unread badge disappears, but the message stays in your head for the rest of the day.

Writing is rarely difficult because we cannot make sentences. It is difficult because a reply has to hold several things at once: what the other person said, what we know about them, what we promised, what tone the relationship can carry and what we are actually prepared to do next.

Copying the conversation into a chatbot creates a second job. It also may expose a conversation that was never yours to upload.

DT Life’s drafting tool starts where the reply already lives.

The context is already on the screen

When you ask it to generate a reply, DT Life captures the relevant native window and uses a local vision model to understand the visible conversation. It looks for the speakers, the person you are replying to and the editable field. It can then recall applicable facts and writing preferences before producing a draft.

If the automatic instruction is not enough, manual mode lets you add the missing intent: “Decline, but keep the door open,” “Tell her Friday is realistic,” or “Make this warmer without making another promise.”

The draft appears for review. You can edit it. DT Life also returns confidence information instead of pretending that every part of the screen was perfectly understood. Only when you accept does it place the final text into the field.

You still press Send.

That boundary is deliberate. A tool that writes in your voice should not quietly become a tool that speaks on your behalf.

It learns from the message you chose

The most useful signal is not the first draft. It is the version you were willing to send.

DT Life records the exact accepted text so it can learn appropriate style over time. If you consistently remove long introductions, prefer contractions or speak differently to a close colleague than to a new client, those accepted edits can improve later drafts.

It does not need to treat every word you type and erase as a statement of who you are. Learning happens after acceptance, from the final choice.

This is where drafting and memory support each other. The twin may already know that a person prefers direct answers or that you promised an update today. That context can make the first draft less generic. The finished reply then gives the twin a better example of how you actually communicate.

Who needs this most

This is not only an email feature.

It helps the freelancer who changes tone between clients, the manager moving between difficult conversations, the person writing in a second language and the job seeker answering recruiters all week. It can help anyone for whom the emotional cost of beginning a reply is larger than its word count.

The aim is not to produce more messages. It is to remove the blank-field moment that turns a five-minute task into three days of low-level dread.

Because the screen is interpreted and the draft is generated on your machine, the conversation does not have to be pasted into a cloud AI service. The people in that thread did not consent to becoming training or account data somewhere else. Local processing respects that fact.

A draft is still a draft

Screen understanding can be wrong. A cropped conversation can hide an important sentence. A model can confuse a speaker, miss sarcasm or suggest a commitment you should not make. A draft that sounds like you is not necessarily a draft that is true.

Review matters most when the message has legal, financial, medical or professional consequences. DT Life is an assistant at the keyboard, not an authority over the situation.

There will also be interfaces it cannot reliably read or fields it cannot inject into. In those cases, copying the reviewed draft yourself is better than pretending automation worked.

The success condition is modest: the difficult first version exists, the uncertainty is visible and you remain the person who decides what leaves the room.

See DT Life’s reply drafting →