Notifications ON Survey - Agents That Need Notifications Cannot Plan Their Own Work
Notifications ON Survey - Agents That Need Notifications Cannot Plan Their Own Work
We surveyed how AI agents handle notifications and found a pattern: agents that depend on external notifications to trigger actions are fundamentally unable to plan their own work. They are reactive systems dressed up as autonomous agents.
The Notification Dependency Trap
A typical notification-driven agent workflow looks like:
- Email arrives - agent processes it
- Calendar reminder fires - agent prepares materials
- Slack message appears - agent drafts a response
- File change detected - agent runs a build
Each action is triggered by an external event. The agent never initiates anything on its own. It is a sophisticated event handler, not an autonomous system.
Why This Matters
An agent that cannot plan its own work cannot:
- Anticipate needs - it waits until you need something before starting to prepare it
- Batch efficiently - it handles each notification as it arrives, even when grouping them would be smarter
- Prioritize across domains - it responds to notifications in order of arrival, not importance
- Do proactive work - it never thinks "the weekly report is due tomorrow, I should start gathering data now"
Moving Beyond Reactive
Agents that plan their own work need:
- A task model - understanding of ongoing projects and their timelines
- Initiative triggers - internal signals that say "now would be a good time to do X" independent of external notifications
- Priority awareness - the ability to say "I know a notification just arrived but what I am doing right now is more important"
- Scheduling capability - not just responding to calendar events but creating and managing work blocks
The difference between a reactive agent and a planning agent is the difference between a thermostat and an architect. Both are useful, but only one can design something new.
Fazm is an open source macOS AI agent. Open source on GitHub.