What makes an assistant “proactive”
It has to see your work continuously, not on request. An assistant that only reads your inbox when you ask it to can’t flag the thread you forgot. You’d have to remember it first. It has to carry context across time, connecting what happened last week to what is happening now, since almost nothing worth surfacing is visible in a single message. And it has to decide on its own initiative that something crosses the bar for your attention.
Drop any one of them and you get something less useful: a chatbot with connectors, a notification feed, or a dashboard nobody opens.
Proactive vs. reactive: what actually changes
| Reactive assistant | Proactive assistant | |
|---|---|---|
| What starts the work | You open it and type a prompt | It notices a change in your tools |
| What it produces | An answer you then go and use | A proposed action you approve |
| Failure mode | You forget to ask | It surfaces things that don’t matter |
| Best at | The task in front of you | The task you haven’t noticed yet |
How a proactive assistant actually works
Connect
You authorise the tools your work lives in (email, chat, files, CRM, tickets) through each one’s own permission flow. It’s the step that makes everything after it possible.
Watch
The assistant keeps a standing view of those tools instead of fetching on demand. It sees threads go quiet, tickets age, deals stall and documents sit unapproved, because it’s looking when nobody asked it to.
Decide what matters
It sorts what it has seen into the things that need you and the things that don’t. Most of what changes in a working day is routine, so this step is about picking out the few items worth raising and leaving the rest alone.
Propose
It tells you what it plans to do, such as the reply it would send or the escalation it would raise, and waits for you to say yes. You don’t have to be the one who remembered the work was outstanding.
Act, once you approve
You accept, and it executes in the underlying tool. The approval step is what keeps you safe: nothing happens in your name until you have seen it, so the assistant can’t do something it wasn’t supposed to.
Why so few assistants are actually proactive
The model isn’t the hard part. Deciding what to surface is. An assistant that raises everything is a worse inbox, and people abandon it within days, so most of the product is the judgment about what to stay quiet about.
And cost. Watching everything, all the time, takes far more inference than answering a question when asked. Run every check at frontier-model rates and you’ll cap the spend before the tool proves itself. Routing cheap work to cheap models isn’t a nice-to-have here. It’s what makes always-on affordable.
Routed to an open model
What to look for in one
- It reads your tools, not just your prompts
- It has a precision story
- It asks before it acts
- It controls what it costs to keep watching
How Hop does it
Hop runs agents in the background across your email and chat. Switch on the built-in ones or describe a new one in plain English, and it watches for the pattern you care about: a thread awaiting a reply that never came, a question nobody answered, the follow-up you said you’d send on Friday.
What comes back isn’t a notification. Hop has already pulled the context together and drafted the reply, so all that’s left for you is to approve it.
Underneath, each task is routed to the model it’s worth: open models for the routine passes, frontier ones for the hard reasoning, which is what makes watching all day affordable. And if you’d like it, we do the setup with you rather than sending a docs link, whatever the size of your team, because deciding what gets surfaced is the part that takes real time.
Two ways in from here: what changed in the category this year, or what this looks like for a sales team.
Today
In 2 hours · 1:00 – 1:30 PM
CloudRelay renewal call
Frequently asked questions
- What is a proactive AI assistant?
- A proactive AI assistant is software that surfaces work you should act on without being asked. It reads the places work is discussed, your email and chat, and notices when something needs attention, such as a follow-up nobody sent or a thread that has gone quiet, then brings it to you with the groundwork already done and a next step attached.
- How is a proactive AI assistant different from a chatbot?
- The trigger. A chatbot is started by a human typing; a proactive assistant is started by a change in your data. A chatbot’s usefulness is capped by your own attention, because you have to think of the question first. That is also why the two work well together: one for thinking and writing, one for making sure nothing gets dropped.
- Is a proactive AI assistant the same as an AI agent?
- No, though they overlap. An agent is defined by how it works: it plans, uses tools, takes several steps. A proactive assistant is defined by what starts it. Most proactive assistants use agent techniques underneath, but plenty of agents sit idle until someone gives them a task.
- Will a proactive assistant spam me with notifications?
- That’s the central risk of the category, and it depends on how well the tool decides what not to raise. An assistant that surfaces everything is a worse inbox, and people abandon it within days. Spend your evaluation on precision rather than capability: ask what it stays silent about, and how fast you can tell it to stop.
- Does a proactive AI assistant act without my permission?
- A well-designed one doesn’t. The standard shape is propose-then-approve: it does the noticing and the preparation, then waits. Most products will let you grant autonomy for narrow, low-stakes actions, but that should be something you switch on.
- What tools does a proactive AI assistant need access to?
- As much of where the work lives as you are comfortable granting, because coverage is what lets it connect the dots. Usually that means email, team chat, document storage, and whichever system holds customers or tickets. Access is granted and revoked per tool through each vendor’s own flow, so you can start with one or two and widen later.
Keep reading
AI personal assistant
What an AI personal assistant is, the four kinds of them, and what they still can’t do.
Read it →AI assistant for sales teams
Call prep, follow-ups and CRM entries, brought to the rep without being asked.
Read it →Proactive AI: what it is and why to try it
What proactive AI means, how it differs from an AI agent, and where the big tools stand.
Read it →Stop asking your assistant what needs doing.
Let it tell you.
Connect your email and chat and Hop starts working out what you’re on the same day.