What is a proactive AI assistant?
A proactive AI assistant notices what needs your attention and tells you first, instead of waiting for you to open it and ask. The good ones stay quiet most of the time, ask before they act, and let you see what they remember.
- GravelYour 6:15 to Denver is delayed 2 hours. There’s a 4:40 on United that still gets you in before dinner. Switch?
- Youdo it
- GravelSwitched. Seat 14C, aisle like you like. Your new boarding pass is in your email.
What does proactive mean for an AI assistant?
Proactive means the assistant starts the conversation. It reads the sources you connect, such as your email and calendar, notices something that needs you, and messages you without being asked: a meeting that moved, a flight that is delayed, a bill that went up, a reply someone is waiting for.
A reactive assistant does the opposite. It sits idle until you open it and type, and it knows only what you tell it in that conversation. Most chatbots work this way. That is fine for questions, but it leaves you responsible for remembering to ask, and the things that cost you the most are usually the ones you did not know to ask about.
How is a proactive AI assistant different from a chatbot?
A chatbot answers; a proactive assistant watches and tells. The differences come down to five things.
| Reactive chatbot | Proactive assistant | |
|---|---|---|
| Who starts | You | The assistant |
| What it knows | What you type or paste | The email and calendar you connect |
| Where you use it | Usually a separate app or website | Can live in your messaging app |
| What it does | Answers and drafts | Notices, proposes, and handles errands after your yes |
| Main risk | Nothing happens if you forget to ask | Too many messages, or acting on a wrong guess |
The last row matters most. Because a proactive assistant speaks first, its quality depends as much on what it chooses not to say as on what it does.
What makes a proactive AI assistant good?
A good one sends signal instead of noise, asks before it acts, keeps quiet hours, and keeps a memory you can read and edit. Four tests tell you most of what you need to know.
Signal over noise
A proactive assistant earns your attention one message at a time. It should weigh how urgent and important something is and how sure it is of the facts, and say nothing when the answer is no. If you ignore its messages, it should send fewer, not more. Look for a daily limit on unprompted messages and a way to mute a topic.
It asks before it acts
Reading is low risk. Sending, canceling and paying are not. A good assistant tells you exactly what it will do and waits for your yes, and it shows a multi-step plan once instead of asking three times for one errand. How much an assistant should do without asking covers the settings in detail.
Quiet hours
It should not text you at 2 AM about something that can wait. Quiet hours you can change, plus one short morning summary for everything that is not urgent, make a proactive assistant livable.
Memory you control
It will remember your preferences, people and plans, so you need to see what it remembers. You should be able to ask what it knows about you, correct it and tell it to forget. If you cannot, that is a warning sign.
What can a proactive AI assistant do for you?
Mostly small things that add up: it catches the change you would have missed and offers to fix it. Typical examples:
- A flight is delayed, and it offers an earlier one. See flight delays.
- Your internet bill went up, and it offers to ask for a better rate. See bills and subscriptions.
- Someone is waiting for your reply, and it drafts one. See email follow-ups.
- A meeting moved and now clashes with another, and it offers to reschedule. See calendar conflicts.
The pattern is the same every time: it notices, tells you in one message, and acts after your yes.
What are the risks of a proactive AI assistant?
The two real risks are access and misjudgment. To notice things, a proactive assistant has to read your email and calendar, so you are trusting it with a lot. And it can be wrong: it can misread a message, or decide something matters when it does not.
Both are manageable if you check a few things first. Ask what it reads and keeps, who at the company can look, how deletion works, and whether it starts with read-only access. The privacy checklist goes through each question. For misjudgment, the best protection is the one above: it shows you what it will do before it does it, so a wrong guess costs you a glance, not an error.
Who is a proactive AI assistant for?
It pays off when your life runs through email and a calendar: appointments, bills, school forms, deliveries, landlords, work threads. In that kind of week there is always something that moved, expired or went unanswered, and a proactive assistant is the one that notices.
It pays off less if you get a handful of emails a week and keep a clear head about them. A simple test: think back over the last few months and count the times a missed email, a changed meeting or a forgotten deadline cost you time or money. If it happened more than once, an assistant that watches for you is worth trying. If you would rather fix it yourself first, how to stop missing important emails covers the habits that help.
How does Gravel work as a proactive assistant?
Gravel is a proactive assistant that lives in your messaging apps: Telegram and WhatsApp during early access, with iMessage coming. It reads the email and calendars you connect, texts you first when something needs you, and handles it when you say yes. It keeps quiet hours from 10 PM to 7:30 AM by default, sends at most a few unprompted messages a day, and puts anything that can wait into a short morning brief.
You choose how much it may do on its own, it starts read-only, and you can ask it what it knows about you and tell it to forget. Text “pause” and all background work stops. Gravel is in early access, with one plan at a planned €12 a month. See how it works, or how Gravel compares with other assistants.
Questions
What is the difference between a proactive and a reactive AI assistant?
A reactive assistant waits until you ask. A proactive assistant watches the sources you connect, such as email and calendar, and messages you first when something needs your attention. Many proactive assistants also carry out the task after you approve it.
Will a proactive AI assistant message me all day?
A good one does not. It should have quiet hours, a limit on unprompted messages and a way to mute topics. Gravel is quiet from 10 PM to 7:30 AM by default, sends at most a few unprompted messages a day, and sends fewer if you ignore them.
Does a proactive AI assistant need access to my email?
To notice things on its own, yes, because it has to read something. Without a connected mailbox or calendar, an assistant has nothing to watch and can only answer questions. Choose one that starts with read-only access and lets you take access back.
Can a proactive AI assistant act without asking me?
That depends on its settings. With Gravel you pick, for each kind of action, one of four levels: just do it, only when I ask, ask me first, or hand it to me. Gravel starts read-only.
Is there a proactive AI assistant for iMessage or WhatsApp?
Gravel is coming to iMessage, but today it works in Telegram and WhatsApp. Gravel is a proactive AI assistant that texts you first in your messaging app when something in your email or calendar needs you. It is in early access.
Keep reading
Updated
Hand Gravel
the small stuff.
We’re opening Gravel in small groups. Leave your email and we’ll let you know when it’s your turn.