There is no single path to learning about AI. But one way to think about AI is through the lens of entrepreneurship. By “entrepreneur,” I don’t just mean starting a business. Instead, I’m referring to a mindset—a way of identifying opportunities, taking risks, and creating value.
Q. As an entrepreneur, where are the opportunities in AI?
A. Agentic AI Workflows
This post is an introduction to Agentic AI Workflows. The good news? You’re already familiar with the concept, even if you know nothing about AI.
Writing an Essay: Zero-Shot vs. Recursive Composition
We all know how to write essays. Broadly speaking, there are two approaches. The first is the “zero-shot” method—writing an essay from start to finish in one uninterrupted go, without pausing for revision.
Zero-Shot Essay Composition
One remarkable historical example of zero-shot composition is Ulysses S. Grant’s Personal Memoirs. Facing terminal throat cancer and financial ruin, the former president and Civil War general wrote his memoir in a desperate bid to provide for his family. Grant worked through excruciating pain, often unable to speak, using morphine and cocaine water to dull his suffering. Despite these hardships, he completed nearly 336,000 words in less than a year, just days before his death in 1885. His memoir, praised for its directness, clarity, and precision, is considered one of the greatest works in American literature.
But we are not Ulysses S. Grant. Most of us can’t write masterpiece essays in one go. Instead, we were taught with good reason a different approach—recursive composition.
Recursive Composition
Recursive composition embraces the idea that nuance and sophistication develops through iteration. This approach involves multiple passes: examining, critiquing, expanding, and refining initial ideas. Each draft serves as raw material for the next iteration, enabling writers to sharpen their arguments and refine their expression. By layering and deepening thought through these revisions, we gradually transform rough concepts into polished prose. Recursion, unlike simple iteration, involves layered iteration.
Non-Agentic AI vs Agentic AI Workflows
Similarly, AI systems can follow two different paradigms: Non-Agentic AI (“Zero-shot”) and Agentic AI Workflows.
Non-Agentic AI (“Zero-shot”) Workflows
Non-agentic AI mimics the zero-shot essay-writing approach. It’s also how most of us interact with AI systems. You provide a single prompt, and the AI generates an output in one pass. While this approach is straightforward, it fails to fully utilize the AI’s potential. It’s akin to expecting a student to write a polished essay in a single attempt, without drafts or revisions.
Agentic AI Workflows
Agentic AI Workflows, by contrast, mirror recursive composition. They involve iterative processes where tasks are refined over multiple cycles. For example, in essay writing, an AI system might start with an outline, gather relevant information, produce a draft, and then engage in cycles of critique and revision. Each iteration builds upon the last, gradually improving the quality of the output.
Why Iteration Matters
The benefits of recursive approaches are evident in real-world applications. Research shows that enabling AI systems to iterate can significantly boost performance. For instance, in coding tasks, recursive workflows have increased accuracy from 67% to 95%. This leap underscores the power of iteration in tackling complex challenges. By mimicking the human process of planning, drafting, and refining, AI systems can deliver more sophisticated and reliable results.
Agentic AI Workflow is emerging as the dominant pattern for designing, building, and optimizing AI Systems.
The Big Picture
The shift from non-agentic to agentic AI workflows reflects our own evolution in mastering complex tasks. Just as we moved from zero-shot writing to recursive composition, AI systems can be designed to iterate and improve.
In the next post, we’ll delve into the AI stack—the layered technologies and processes that enable these sophisticated workflows. For entrepreneurs, understanding the stack is essential to identifying opportunities in the rapidly changing AI landscape.