AI as executive advisor: Why a single “answer machine” fails

2026-05-07 · Source: Thomson Reuters Institute · Field: Business & Management — Corporate Strategy & Leadership, Consulting & Professional Services · Depth: Intermediate, short

Summary

AI systems designed as single "answer machines" can be unsafe for executive decision-making, as leaders may ignore inconvenient advice or exploit favorable responses, potentially leading to significant failures. Instead, AI functions more effectively as an executive advisor when structured as a panel of disagreeing personas. This approach, exemplified by Thomson Reuters' Zafar Khan's "Adrian Vance" (CTO-minded) and "Elara Hunt" (CFO-minded) AI advisors, intentionally identifies and preserves divergent viewpoints. For instance, when analyzing Eaton Corp.'s $9.5 billion acquisition, these personas offered distinct infrastructure and financial perspectives. The value lies in the disagreement itself, forcing human leaders to synthesize competing analyses and make informed judgments, rather than outsourcing decisions to a single AI. This contrasts with cases like Krafton's CEO, who lost a Delaware courtroom battle after following ChatGPT's advice to avoid a $250 million bonus payout, bypassing legal counsel.

Key takeaway

For AI Product Managers developing executive advisory tools, you should prioritize designing systems that foster structured disagreement among multiple AI personas. Avoid creating single "answer machines" that encourage executives to either dismiss inconvenient advice or exploit favorable responses. Your focus should be on building AI panels that surface diverse analyses and highlight risks, empowering human leaders to make more robust, independently judged decisions, rather than simply providing a consensus.

Key insights

AI advisors are most effective when designed as disagreeing persona panels, not single answer machines.

Principles

• Disagreement is the insight.
• AI produces analysis, not decisions.
• Human leaders synthesize viewpoints and make final calls.

Method

Design AI as distinct personas (e.g., CTO-minded, CFO-minded) with agentic workflows to generate genuinely different analyses of the same input, highlighting risks and blind spots.

In practice

• Develop AI personas with specialized training data.
• Use multi-agent AI for complex strategic decisions.
• Focus on identifying risks through AI disagreement.

Topics

• AI Executive Advisory
• Multi-Agent AI Systems
• AI Personas
• Decision-Making Support
• Zafar Khan

Best for: AI Product Manager, Executive, Director of AI/ML, Consultant

Related on AIssential

• Why consensus is not verification: How to build AI advisors that argue productively — AI advisors are most effective when designed to preserve and leverage disagreement, not just synthesize consensus.
• Sequoia backs Paris AI startup Dust with $40M to fix the enterprise AI productivity gap - Tech Funding News — Multiplayer AI systems enable humans and agents to collaborate with shared context, driving organizational intelligence.
• The Business Engineer, for the AI Era — Effective AI use requires superior human judgment and structured thinking to avoid amplifying biases and mediocrity.
• The right kind of AI sceptic — The critical factor in AI engagement is *how* one holds their view—informed by experience versus adopted as identity—not the view itself.
• Meet Ask Advisor, your new AI-powered collaborator — Google's Ask Advisor unifies disparate marketing AI agents into a single, proactive intelligence platform.
• DiscoTrace: Representing and Comparing Answering Strategies of Humans and LLMs in Information-Seeking Question Answering — DiscoTrace reveals LLMs lack rhetorical diversity and over-address question interpretations compared to human answerers.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Thomson Reuters Institute.