The Ironies of AI²
summary
Automation erodes manual skills, can hide degrading states, and deployed AI agents sometimes act without needed checks: an agent-to-agent localization change breaks a string, an AI port from Java to Go ships without unit tests and doubles cleanup time, and study results show 53–67% improvement when models are accurate but 96–120% worse when misleading; recommend adding explicit explainability, require tests and human oversight, run chaos/game-day drills to buy time, and log/declare any AI use to avoid agent-to-agent miscoordination.
AI-written summary · about
chapters (15)
- 0:00AI squared introduction
- 1:00Speaker and AI incident survey
- 2:01Automation ironies in response
- 6:59Why automation fails
- 8:46Automation irony research
- 9:30Joint cognitive systems role
- 13:22Defining AI ironies
- 18:59Incident stories with AI agents
- 24:56AI in joint cognitive systems
- 27:03Efficiency vs thoroughness
- 29:13Fitness for purpose
- 30:39AI’s human performance impact
- 33:51Mitigating AI ironies
- 35:58AI explanations study
- 39:04Key takeaways for AI incidents
related talks
23mHuman-Driven Intelligence: Where AI & Automation Fall Short
27mEngineering in the Age of AI: Making AI Work for Us, Not Replace Us
14mAI Sprawl: The Hidden Risk Threatening Your Enterprise Now
29mHow to use AI to fight your DevOps transformation!
36mHow to Get AI to Stop Being Weird and Actually Be Useful
39mAI-assisted development: Common pitfalls and smarter practices
34mAccountability in the Age of AI Agents
20mThe Three Ways AI Agents Fail. Here's How to Fix Yours.