Stop Prompting AI. Start Directing It
James Yang/theispot.com The Research The authors drew on two streams of research for this article. The first was a qualitative study of AI-assisted discovery that identified four pathways by which AI generates surprising insights during analytical work.i The second stream was an examination of how organizations function as algorithmic assemblages; it showed that algorithmic systems […]
The authors have conducted research on AI-assisted discovery, focusing on four pathways through which AI generates surprising insights during analytical work. The second stream of research examined how organizations function as algorithmic assemblages, highlighting the impact of data accessibility, configured capabilities, and distributed agency in shaping these systems.
The key takeaway is that the most valuable output from professionals is insight, a genuinely new way of understanding a problem or recognizing patterns that others have not identified. This insight is difficult to achieve due to the expert knowledge that often limits the parameters of what professionals look for. Conversational AI, when prompted, can help surface competing interpretations and challenge assumptions.
However, agentic AI, which is configured and directed rather than conversed with, can delve deeper. Unlike prompted conversations with their limitations, agentic systems have access to more data and maintain analytical orientations across entire data sets, leading to deeper insights. This process requires a unique professional skill: directing intelligence.
Two different ways of working with AI include the traditional method of prompting, where a human supplies context and guides the analytical process, and the more advanced method of agentic AI, where the user configures the system to operate proactively, accessing persistent context, specialized capabilities, and analytical orientations to reveal different patterns and insights.
Written by urgent.news from MIT Sloan's reporting — not their text. Machine-written — it may contain errors, so check the original before relying on it.
