Daniel Dines, chief executive of the robotic process automation specialist UiPath, has finished a 168-page book titled The Work That Remains. The work was produced with the assistance of the language model ChatGPT and a method he calls "adversarial prompts".
How the book was created
Dines told us the writing process resembled a Socratic dialogue. He would feed the AI a theory, ask it to read a supporting article, then request an opposing view. The model would generate extensive text, at one stage the draft reached almost 100,000 words, and Dines trimmed it down to the final 40,000-word manuscript.
It was really like a Socratic dialogue, with each session involving a series of hypotheses, challenges and instructions.
The iterative cycle of proposing, challenging and refining ideas took three years, during which Dines also ran UiPath's global operations. He said the book's concepts and the company's strategy evolved together, feeding each other.
Why the experiment matters
The book illustrates a model for future collaboration between humans and artificial intelligence: AI proposes, humans decide, and automation executes. Dines highlighted two limits of the technology. First, the AI's output is only as good as the data it has been trained on, making it strong on known problems but weak on truly novel ideas. Second, without critical feedback the system can produce "hallucinations" or become stuck in unproductive loops.
By confronting the model with opposing arguments, Dines believes he reduced these risks and demonstrated that thoughtful partnership can turn raw AI output into a coherent, public-facing narrative.
What comes next
Dines plans to use the book's framework to guide UiPath's own product development, encouraging customers to view AI as a collaborative partner rather than a replacement. He also hopes the experience will inspire other leaders to experiment with similar "adversarial prompting" techniques, especially as large-language models become more widely available.
For readers interested in the full text, the PDF is available for download on UiPath's website.