← All lessons
Day 84 of 90

Disaster Recovery for Agent-Dependent Commerce

8 min read

Key Concepts

  • 1When commerce depends on an agent, an agent-stack failure can halt sales, so disaster recovery must plan for LLM outages, tool failures, and data-store loss
  • 2Provider dependency is a single point of failure: an LLM provider outage can take the agent down, so fallbacks (alternate provider, degraded mode) are essential
  • 3Graceful degradation to a non-agent path (traditional search/checkout) keeps commerce flowing when the agent is unavailable — the store shouldn't die with the agent
  • 4State and memory recovery matter: losing conversation state, carts, or long-term memory damages trust and revenue, so backups and durability are required
  • 5RTO/RPO thinking applies: define acceptable recovery time and data-loss windows for agent components based on their revenue criticality
  • 6Failover must be tested, not assumed: a fallback that's never exercised often fails when actually needed, so DR drills matter
  • 7Communication and honest degraded-mode messaging preserve trust during an incident better than a broken, confidently-wrong agent

Why It Matters for PMs

Business continuity: If the agent is a primary sales channel, its failure halts revenue. DR planning for agent components is a business-continuity necessity.
Fallback to non-agent commerce: A degraded traditional path keeps the store selling when the agent is down — the store shouldn't die with the agent.
Tested failover: Untested DR usually fails when needed. Drills and honest degraded-mode communication protect revenue and trust during real incidents.

Test Your Understanding

Question 1 (Multiple Choice)

Why is LLM provider dependency a disaster-recovery concern?

Question 2 (Multiple Choice)

What keeps commerce flowing when the agent is unavailable?

Question 3 (Multiple Choice)

Why must failover be tested rather than assumed?

Question 4 (Short Answer)

How would you design a fallback so an LLM provider outage doesn't halt sales?

Think about your answer before revealing the explanation.

Question 5 (Short Answer)

Why do RTO/RPO concepts apply to agent components, and how would you set them?

Think about your answer before revealing the explanation.