The short answer
Proactive AI is software that starts the work instead of waiting to be asked. It reads the tools where work already happens, like email and messaging apps, then works out what a person is in the middle of and brings them the next step with the action attached. The difference from a chatbot is who notices the work. A chatbot needs you to spot it first. A proactive assistant spots it for you, then asks for approval before acting.
Proactive AI describes an assistant that starts a task on its own rather than waiting for you to ask. Three of the biggest launches this year gesture at the idea, and only one of them actually does it. That gap is worth understanding before you buy anything.
What is proactive AI?
Every assistant you have used waits. You notice something needs doing, you open a tab, you describe it, and you get an answer back. That last part got very good over the past three years. The first part did not change at all. You are still the one who has to notice.
Proactive AI moves that step. It watches the places where work gets discussed, usually email and messaging apps, works out what you are in the middle of, and brings you the next thing with a draft already attached. You approve it or you bin it.
The word is doing a lot of work in marketing right now, so it helps to be precise. Running on a schedule is not proactive. Firing on a webhook is not proactive. Both are automation, both are genuinely useful, and both have existed for fifteen years under other names. Proactive means the software worked out that there was a task at all.
How it is different: three generations of assistant
It is easier to see if you lay out how we got here. Each generation took a different job off your plate.
Reactive, the chatbot. You ask, it answers. Genuinely useful, and it only ever helps with what is already on your list. Nothing gets onto that list by itself.
Scheduled, the agent. It runs without you watching. Every Monday, or when a ticket closes, or at six in the morning before a client call. This is real progress and most of the good tooling shipped in the last two years lives here. But you had to imagine the task and describe it in advance. The work did not disappear. It moved from doing to specifying.
Proactive. It reads your tools and works out the task itself. Nobody noticed it and nobody wrote it down, and it still arrives.
The first generation removed the doing. The second removed the remembering. The third removes the noticing, which is the one that quietly eats a week. Most of what actually goes wrong at work is not a task somebody failed to complete. It is a task nobody spotted: the commitment made in a thread on Tuesday, the customer question nobody picked up, the renewal everyone assumed was handled.
| Generation | Who notices the work | Who describes the task | Who does it |
|---|---|---|---|
| Chatbot | You | You | AI |
| Agent | You | You, in advance | AI |
| Proactive AI | AI | AI | AI, once you approve |
Who is actually building proactive AI
Three launches this year get filed under the same heading, and they are not the same thing.
ChatGPT Work arrived in July 2026. It is an agent that will stay on a job for hours and hand back finished spreadsheets, decks and documents, pulling context from the tools you connect to it. It is good, and it is a real step past the chat box. It still starts with you.
Claude Cowork shipped in January 2026 and reached web and mobile in July. Same shape, with scheduling attached. Tell it to have Monday’s client prep ready at six in the morning and it works through the email threads and transcripts and leaves the follow-up drafted. Useful. You still had to think of Monday’s client prep.
Microsoft Scout was announced at Build in June 2026, and this is the one that moves. Microsoft describes it as always on, staying active in the background and taking action without needing to be prompted each time. Sensitive actions can require a person to sign off. It is in private preview.
So of this year’s three big launches, two are agents and one is proactive, and the proactive one is not something most teams can get yet.
What we are building at Hop
Our aim is that you never have to write an extra word. Not one. If Hop is doing its job the work is already in front of you with the next step attached, and prompting it would be the slow way round.
We charge by credits. We are still trying to get your work done with minimal to and fro between you and the AI. Those two things are in tension and we have picked a side.
That is an odd thing to optimise for when the pricing is usage based. Every prompt you do not write is one we do not bill. We think it is the right trade anyway, because an assistant that needs you to notice the work first has not removed the expensive part of the job. It has made the cheap part faster.
So Hop reads the places where work actually gets discussed, email and messaging apps, alongside the systems holding the context, and brings you what needs doing with the action already on it. It proposes and you approve. And of course it will miss things, which is why you can still prompt it directly whenever you want.
For teams, we connect it to your tools and do the setup for you ourselves rather than sending you a documentation link, because that part is most of the project and it is where rollouts usually die.
If you want the longer version of how it works, that is on Hop as a proactive AI assistant. If you are weighing the field rather than us, we compared popular tools and said where each one falls short, ours included.