what survives when a long-running agent task gets interrupted?
Summary
Explores what state or progress is retained when a long-running AI agent task is interrupted, addressing implications for reliability and recovery.
Similar Articles
How are you handling recovery when AI agents fail mid-task in production? and How often this happens for you?
A discussion query asking developers how they handle recovery when AI agents crash mid-task in production, exploring approaches like restarting, persisting state, using checkpoints, or manual inspection.
Recoverability as a System Primitive for Long-Horizon AI Agents
Introduces recoverability as a system primitive for long-horizon AI agents, emphasizing evidence-based recovery decisions to maintain justified progress during task interruptions.
What if continuity in long-running AI agents is not about preserving state, but preserving a trajectory through change?
The article questions whether continuity in long-running AI agents should focus on preserving state or maintaining coherence through inevitable change, exploring implications for memory, actions, and interactions.
What should happen when a user changes the task while an agent job is still running?
The article explores how AI agents should handle user-initiated task changes during execution, proposing solutions like task revision IDs and discussing edge cases to avoid stale or conflicting actions.
Long-running AI agents may have a bigger continuity problem than memory
Reflects on the continuity problem for long-running AI agents, arguing that a deterministic control layer is needed to manage authoritative state, and questions whether existing infrastructure like IAM, transactions, and provenance is sufficient.