This Week's Term: Proactive AI agents - AI agents that start the interaction themselves. They watch your context, such as your inbox, your calendar or an upcoming trip, and act or reach out to you before you ask, within limits you set.
The AI assistants many of us have used so far are reactive. They wait for a prompt, answer it, and stop. The personal agents launched in the last two months work differently. Instinct texts or calls its users first when something needs doing, and OpenAI describes its new dots as agents that keep working toward your goals while you are away. I review these agents in this issue's tools post, Instinct and the new personal agents.
At World Summit AI in Amsterdam (7-8 October 2026), Xavier Amatriain (Chief AI & Data Officer at Expedia Group) pointed to proactivity during a trip, and not only in the planning before it: an agent that takes care of the needs and wishes to change or add things as you travel. He also saw a big opportunity in group itineraries that consider the needs of each member of the group.
The business implications
For business leaders, proactive agents may show up in two places. Your customers' agents will start contacting you to book and negotiate on their owners' behalf. And your own service can start reaching out to customers first, before they notice a problem or ask for a change.
A proactive agent also raises a practical design question: when may it act without asking? OpenAI's answer for dots is a capability called Custom Rules, which lets you allow specific actions, require approval for them, or block them.
Recommended resource
In my library of AI agent design patterns, this question is covered by a pattern called the autonomy dial: deciding separately for each type of task how much the agent may do on its own.
Your action step
If you are designing a proactive service for your customers, I suggest you start with that question. List the types of tasks the agent would do, and for each one decide how much it may do without asking. Then give people a simple way to change the answer later.
If you are designing agents that act for your customers or your teams, that is the kind of question I work through in AI strategy advisory engagements and in sessions as an AI keynote speaker across Europe.