In her November 21 essay for The AI Disruptor Playbook, Wei Manfredi uses recent technology outages to make a point about AI: the outages were not caused by agents, but they show how tightly connected systems can turn a small failure into a business-wide disruption. Agents add more paths for that to happen because they can decide which systems to call, whether to retry, and what to do next.
What Manfredi recommends
- Keep one failure from spreading. Identify shared dependencies and give each critical workflow a fallback, such as operating with limited features, safely queuing work, or handing it to a person.
- Design for work that stops halfway through. Record what actually happened before retrying. If a hotel payment times out after charging a guest, the next step is to check whether the charge went through, not to charge again. The same record should let a reservations agent pick up where the automated workflow stopped.
- Earn autonomy where mistakes are costly. A tool that answers staff questions can be tested with fewer controls than one that issues refunds or changes inventory. For consequential actions, limit retries, set conditions that stop the agent, and test that a person can take over.
A starting point for hotel teams
Manfredi proposes a 90-day plan: first map dependencies and assign owners for critical workflows like booking, check-in, and refunds. Then put fallbacks and recovery steps in place. Finally, interrupt a dependency in a controlled environment and check whether the failure stays contained, transactions remain consistent, and staff can resume the work. Track how far failures spread, how long safe recovery takes, and how many workflows have tested fallbacks.