Agentic AI Interview Questions
Dashboard
0%
1
Curious builder0 XP earned · 300 to level 2
0 daysFinish a lesson to begin
Badge collection0 of 6 unlocked
51 small wins to finish your pathNext question →

Q24IntermediateConcept

Why do agents need state persistence and checkpointing?

30-second answerSay your answer out loud first, then reveal.
Each agent step is followed by a checkpoint; a crash in step 2 falls back to the last checkpoint, and step 3 is interrupted to wait hours for a human before resuming from the checkpoint into step 4.

What a checkpoint contains: messages, the agent's structured state (plan, variables), which node or step comes next, pending tool calls, and metadata.

What it enables

  1. Fault tolerance: a server restart or deploy doesn't kill in-flight tasks.
  2. Human-in-the-loop: pause indefinitely, then resume (Q11).
  3. Time travel / debugging: reload the state at step 7, change something, and re-run from there.
  4. Multi-turn threads: the conversation continues across requests on stateless servers.

Implementation options

  • Agent frameworks: LangGraph checkpointers (Postgres, SQLite, Redis), with a thread ID per conversation.
  • Durable execution engines: Temporal, Inngest, AWS Step Functions, or similar. These make each step a recorded activity that is replayed after failure.

Subtle point: resuming is only safe if tool calls are idempotent or their completion was recorded. Otherwise a resumed step may send the email twice (Q32).

PreviousWhat does observability look like for agents, and why is it harder than for normal apps?

This is what real progress feels like.