August 26, 2024
Inversion of the AI agent model — long-running processes that pause for human direction when necessary ...
The demo Dex of @humanlayer_dev did at @AITinkerers in SF last week really stuck with me.
He talked about a few things, including:
1. How long-running agent processes can unlock new use cases.
2. For long-running processes, you might need to invert the typical agent model.
In most agent frameworks, the agent runs in response to a human action. But for long-running processes, you will often ask humans to "run" on behalf of an agent. For example, a human might need to approve an action, or provide more context in the middle of a multi-step process.
These two ideas map nicely onto some of the voice AI things we've helped customers build at @trydaily. For example, in healthcare and elder care contexts, agents might make phone calls to people to gather information or to remind them to do something.
I like these two "tools to think with" for building AI agent systems: engineering long-running processes, and building hooks for human-in-the-loop actions.