Replanning
Throwing out or reworking the plan mid-task when reality doesn’t match what the agent expected.
Think of it like
Hitting a road closure and rerouting — the destination’s the same, the path you drew is now useless.
Example
The agent planned to use an API that turns out to be down, so it stops, replans around a web search, and continues.
How it actually works
Plans made up front rarely survive contact with a messy environment, so replanning lets an agent revise its approach when observations contradict its assumptions. It’s what makes agents robust rather than brittle — but it’s a balancing act: replan too eagerly and the agent thrashes, never finishing; too reluctantly and it keeps executing a plan that can’t work. Knowing when the plan is truly broken is the hard part.
For product teams
The difference between an agent that adapts and one that stubbornly runs a doomed plan into the ground.
For engineers
Revising the remaining plan when observations invalidate prior assumptions, balancing stability against adaptivity.
Related
- Planning — What replanning revises.
- Observation — The signals that trigger a rethink.
- Self-Correction — The self-fixing behavior it’s part of.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome