It’s a common scene in the modern enterprise: developers are using AI to crush unit tests, designers are using it to generate assets, and QA is using it to write scripts. On paper, everyone is “using AI.” But in reality, we are often just applying high-tech grease to an old-fashioned machine. We’ve optimized the silos, but we haven’t touched the walls between them.
I realized this after a side project forced me to change my own workflow. I was building an app for an organization and needed a way to bridge the gap between a vague idea and working software without the typical overhead.
The “Back of the Napkin” Breakthrough
Instead of writing a massive requirements document, I had Gemini generate mockups in HTML and CSS. I gave it a specific constraint: make it look like a “back of the napkin” sketch.
The goal was psychological; I wanted the client to focus on the flow and the logic, not the hex codes or the padding. It worked brilliantly for the client, but the real magic happened when I moved to the build phase. I realized that these informal, “messy” screens weren’t just mockups—they were the ultimate requirements document.
Because the AI had helped me build the visual logic, it already “understood” the intent of the application. I fed those screens back into the model, and it gleaned the functional requirements, data structures, and edge cases with startling accuracy. I had a working application in record time, moving straight from logic to final styling with almost zero back-and-forth.
The Corporate Paradox: Efficiency vs. Disruption
This experience highlighted a glaring issue in how many organizations approach AI. At work, we tend to respect the traditional boundaries of the SDLC. We hand off a document from one silo to the next, with each role using AI to speed up their specific slice of the pie.
The premise is simple: Businesses are failing to capture the full value of AI because they are using it to perform tasks as currently defined, rather than using it to collapse the process itself.
If you are just using AI to write code faster, you’re missing the point. The real value is unlocked when the lines between “Requirement,” “Design,” and “Code” blur into a single, continuous stream.
Why Siloed AI is a Half-Measure
When we keep AI trapped within existing roles:
Context is lost in handoffs: The “intent” of a feature gets diluted as it moves from a PRD to a Figma file to a Jira ticket.
The process remains linear: We are still waiting for “Step A” to finish before “Step B” starts, even though AI can often see the end result from the beginning.
Innovation is incremental: We get 20% faster at doing things the old way, rather than 500% faster by doing things a new way.
To realize the true potential of this technology, we have to be willing to be disruptive to the entire process. We shouldn’t be asking “How can AI help this developer?” but rather “How can AI change how we get from an idea to a product?”
The “back of the napkin” approach taught me that the most valuable thing AI can do is remove the friction between thinking and building. If your organization is still treating AI as a personal assistant for individual roles, you aren’t leading an AI transformation—you’re just upgrading your word processors.
Article Written By Norm Murrin