The dangers of deploying AI agents: ADAPT Advisor, Dr. Michael Kollo
Dec 16, 2024 | 3 mins2025 is being heralded as the year of the autonomous AI agent. Tech giants are rolling out generative AI assistants that can make decisions, interpret data, and trigger actions across your organisation.
The pitch is compelling: string these agents together like Christmas lights, and they’ll streamline every business process from market analysis to policy enforcement.
But this is far from the reality of AI agents.
The problem is simple but profound. Generative AI agents make decisions. Each decision builds on previous decisions. And this creates two distinct challenges that most companies aren’t ready for: compounding uncertainty and path dependency.
Consider uncertainty first. Each agent takes a previous agent’s uncertain judgment as fact.
For example:
- Agent One evaluates market data as “potentially concerning.”
- Agent Two treats this as confirmed and elevates it.
- Agent Three, working off two layers of interpretation, finds patterns supporting the concern.
- Agent Four, transforms an initial…