top of page
Search

Why Workflows Should Come After You Learn to Think With AI

This past weekend, I conducted training for a group of business leaders, and we started with a prompt I had designed for instructional purposes.  The question came up almost immediately: “Aren’t we all going to get the same answer?”


Everyone was using the same tool and the same prompt, so it was a fair question. Given how often AI is framed as a matter of finding the “right” prompt, the expectation of consistent output makes sense.


In case anyone reading may not know, they were not identical outputs. 


The responses were similar in direction, but they were variations with different shifts in structure, direction, and even working.  It wasn’t until we were toward the end of the training, that I realized how grateful I was for that moment.  So much so, I may just attempt to replicate it from now on. 


The Problem in Pushing for Systems


As we approached the end of the session the group began asking about automation tools, workflows, and building systems, which we do discuss toward the end of our Level One training. 


I expect these questions from professionals simply because, in many cases, people are being handed workflows or pre-built processes and told they are ready to scale. Think about this for a moment. 


OpenAI tools have only been widely accessible to the general public for a little over three years. In that time, the conversation has moved quickly from basic usage to system-building and automation.

Society has had this tool for less than four years and despite the fact many of them do not even know how to operate it at the most basic level, we are handing them pre-built processes for them to work with and scale.  Sometimes that works—at least in the beginning. But it will not hold up long term.


AI does not return identical outputs. It produces a range of possible responses from the same input.  This is part of the nature of AI.  That variability is part of how the tool works.


The issue shows up when people are asked to rely on systems they do not fully understand. Without the ability to adjust inputs, evaluate responses, and refine direction, the system starts to drift.


The changes are rarely obvious at first. The output still looks acceptable; it just becomes less useful over time.


There is research backing this up from the National Library of Medicine. Their study on human and AI collaboration found that outcomes improve when people stay actively involved in reviewing and refining what the model produces. When that layer of judgment is removed, consistency tends to suffer. [1]


In other words, the system is only as stable as the person guiding it.


Learning Before Building


This is why Structured Thinking training starts in a different place.


We don’t start with an introduction to workflows or systems, we focus on how people are actually working with the tool.


That includes how they structure inputs, how they respond to what comes back, and how they decide whether something is usable. It is less about getting the perfect prompt and more about developing a repeatable way of thinking through a problem.


There is a growing body of work that frames generative AI as part of a learning process rather than a one-time solution. The aim of this research is “to ensure that the quality of learning is truly augmented, and not diminished, by the use of AI-based systems…”.


The study goes on to find that only by training someone how to think and operate the system prior to providing prebuilt frameworks are you truly augmenting with AI.  [2]


When Workflows and Systems Make Sense


When people spend time learning how to think with AI, how to evaluate its work, patterns start to emerge. Certain adjustments lead to better results. Certain steps get repeated, and over time, those steps become more consistent.


That is where workflows organically come from.


They don’t come from templates or pre-built structures, but from repeated use. Science Direct published a study that said, “It is essential for users to interact with the system in a specific manner to elicit the desired responses.”  It goes on to discuss the importance of the human mind moving through the cycles of drafting, revising, and exploring before settling into a process that works for them. [3]


Workflows are the result of that process, not the starting point. Once those patterns are clear, building systems becomes much more effective.


At that stage, there is a better understanding of how to guide the process and maintain quality over time. Variability is still present, but it is no longer a problem. It becomes something that can be managed.


Without that foundation, systems often give the appearance of consistency without actually delivering it.


Why the Order Matters


If AI produced identical outputs every time, prompts would be the main focus. Neither of those statements are true.  AI does not operate that way.


What I am seeing more often is teams trying to standardize outputs before they understand how those outputs are created. Don’t get me wrong, it can work in the short term- not always, but it can. Make no mistake, though it will break down under real use. 


All systems eventually drift and start to break without human intervention, fine-tuning, or maintenance work.


The challenge is when the human intervention is needed, is the knowledge structure in place to adequately intervene? 


If you cannot answer yes to that question about you and your team, if you do not know what to adjust in this situation or that situation to place the workflow or system back on track; “Houston, we have a problem.”


-------------


Comments

bottom of page