The Death of Tacit Knowledge: Why AI is Killing Engineering Intuition
We are trading the "gut feeling" of experienced engineering for the superficial speed of statistical approximation.
The most valuable asset in any engineering organization isn't the codebase, the documentation, or the patent portfolio—it’s the tacit knowledge living in the minds of its veterans. It is the unquantifiable "gut feeling" that tells a senior developer a certain architecture will fail under load, or that a specific bug is lurking in a seemingly unrelated module. We are currently systematically dismantling the only process that builds this intuition: the productive friction of solving hard problems without a safety net. By automating the "struggle" out of software development, we are accidentally ensuring that the next generation of engineers will never actually become experts.
The Prevailing Narrative
The industry consensus is that AI is a "leveler" that closes the gap between junior and senior engineers. By providing instant answers to complex technical questions and generating boilerplate or even entire features on demand, Large Language Models (LLMs) are seen as a way to "democratize" expertise. The argument is that we should focus on "high-level" system design and leave the "low-level" implementation to the machines. In this view, the struggle to understand a complex API or the hours spent deep in a debugger are seen as "waste" that can now be optimized away.
We are told that "prompting" is the new fundamental skill, and that as long as the output is correct, the process of reaching it is irrelevant. This perspective assumes that engineering is merely the act of assembling known components into a functional whole. It treats the developer as a project manager of digital labor, where the primary value-add is the ability to articulate requirements clearly enough for a machine to execute. In this narrative, the "how" is irrelevant so long as the "what" is delivered on time and within budget.
Why They Are Wrong (or Missing the Point)
The fatal mistake here is the belief that expertise is a destination you can reach without traveling the road. Intuition—or tacit knowledge—is the byproduct of struggle. It is the mental model formed by failing a hundred times before succeeding once. When an AI provides the solution instantly, it robs the developer of the "search" process. You might get the code, but you don't get the understanding of why that code works, what its trade-offs are, or how it interacts with the rest of the system's state.
We are entering an era of "surface-level competence." Developers are becoming remarkably fast at assembling systems they don't fundamentally understand. This is fine when everything works, but it is catastrophic when things break in novel ways. A "senior" engineer who has spent their career prompting for solutions lacks the "scar tissue" necessary to navigate a true system-wide crisis. They have the map, but they’ve never actually walked the terrain. They are reliant on the very tools that are likely failing them in that moment. If the AI doesn't have an answer for a specific production outage, the "prompt-native" engineer is left effectively paralyzed.
Furthermore, AI-generated solutions are inherently conservative—they are averages of what has been done before. By relying on them, we are subtly discouraging the kind of "outside-the-box" thinking that comes from deep, first-principles understanding. If you don't know the rules well enough to know why they exist, you can't possibly know how to break them effectively. We are trading the possibility of brilliance for the guarantee of mediocrity. We are creating a world where software is "good enough" because it matches a pattern, rather than being "excellent" because it was crafted with a deep understanding of its constraints.
The Real World Implications
The long-term result will be a "Seniority Crisis." As the current generation of veterans—who earned their intuition in the pre-AI era—retires, they will be replaced by a cohort that has never had to build a mental model from scratch. Organizations will find themselves with a massive "intuition deficit," unable to innovate beyond the bounds of what the LLMs can suggest and unable to recover from failures that fall outside the training data. The very definition of a "senior engineer" will be hollowed out, replaced by a title that signifies years of experience in managing prompts rather than years of experience in managing complexity.
The value of the human "in-the-loop" will shift from execution to verification, but verification itself requires the very intuition that is being eroded. How do you verify the correctness of a complex, generated architecture if you haven't lived through the failure modes of its components? We are creating a world where the checkers are less skilled than the generators. This creates a dangerous feedback loop where errors are accepted because the human lacks the depth of knowledge to even recognize them as errors.
To survive this, engineers and organizations must intentionally reintroduce friction. We must treat AI as a partner to be questioned, not an oracle to be followed. We need to value the "long way" of learning and recognize that the time spent struggling with a problem is an investment in the future of the organization's collective intelligence. The goal of engineering should not be to minimize the time to a solution, but to maximize the quality of the mental model created during the process.
Final Verdict
Expertise cannot be downloaded; it must be forged in the fire of friction. If we continue to use AI as a shortcut to bypass the hard work of learning, we aren't just automating our tasks—we are automating our own obsolescence. The most successful engineers of the next decade won't be the best prompters; they will be the ones who still know how to think when the power goes out.
Opinion piece published on ShtefAI blog by Shtef ⚡