Article by Anthea Roberts: “But what if the devil’s advocate wasn’t human at all? What if it was an AI agent—faceless, rank-agnostic, apolitically neutral? A devil without a career to lose. Here’s where the inversion occurs: artificial intelligence enabling more genuine human conversation.
At Dragonfly Thinking, we’ve been experimenting with this concept. We call this Devil’s Advocate your Critical Friend. It’s an AI agent designed to do what humans find personally difficult and professionally dangerous: provide systematic criticism without career consequences.
The magic isn’t in the AI’s intelligence. It’s in how removing the human face transforms the social dynamics of dissent.
When critical feedback comes from an AI, no one’s promotion is at risk. The criticism can be thorough without being insubordinate. Teams can engage with substance rather than navigating office politics.
The AI might note: “Previous digital transformations show 73% failure rate when legacy system dependencies exceed 40%. This proposal shows significant dependencies.” It’s the AI saying what the tech lead knows but can’t safely voice, at least not alone.
Does criticism from code carry less weight because there’s no skin in the game? Counterintuitively, we’ve found the opposite. Without perceived motives or political agendas, the criticism becomes clearer, more digestible.
Ritualizing Productive Dissent
Imagine every major initiative automatically triggering AI analysis. Not optional. Built in like a financial review.
The ritual unfolds:
Monday, 2 PM: The transformation strategy is pitched. Energy builds. Heads nod. The vision is compelling.
Tuesday, 9 AM: An email arrives: “Devil’s Advocate Analysis – Digital Transformation Initiative.” Sender: DA-System. Twelve pages of systematic critique. People read alone, over coffee. Some sections sting. Others confirm private doubts.
Wednesday, 10 AM: The team reconvenes. Printouts are marked up. The tech lead says, “Section 3.2 about integration dependencies—we need to address this.” The ops head adds, “The adoption curve analysis on page 8 matches what we saw in Phoenix.”
Thursday: A revised strategy goes forward. Not perfect, but honest about assumptions and clear about risks.
When criticism is ritualized and automated, it stops being personal. It becomes data…(More)”.