Coding and Tacit Knowledge in the AI Era

VZ Lens

From a VZ lens, this piece is not for passive trend tracking - it is a strategic decision input. AI can generate syntax, but tacit engineering judgment remains human leverage. Sustainable productivity comes from preserving that invisible layer. Its advantage appears only when converted into concrete operating choices.

TL;DR

Nonaka and Takeuchi’s SECI model describes how tacit knowledge becomes explicit and vice versa. Vibe coding skips the most important step—socialization, where the master passes on what cannot be put into words. Without a master, there is no tacit transfer—and coding turns from a craft into prompt-pushing.

---

A Baker and a Prompt: Where Does Knowledge Begin?

The shore of Lake Tihany, morning fog. The bakery on the shore is already open—the baker rose at three in the morning. The smell of bread fills the entire street. I ask him how he knows when the dough is ready. “From my fingers,” he says. “When it just lets go.”

This moment is the pinnacle of tacit knowledge. He can’t put it into words, but his body, his fingers know. As philosopher Michael Polányi famously put it: “We know more than we can tell.” This tacit knowledge is the foundation of every professional craft, whether baker, surgeon, or software developer.

Nonaka and Takeuchi describe precisely this moment in The Knowledge-Creating Company—literally. Ikuko Tanaka, a software developer, was working on Matsushita’s bread maker project. The machine didn’t bake good bread because it only followed the explicit recipe: how much flour, water, and temperature. Tanaka went to a master baker and, through socialization, learned the rhythm of kneading—not from a book, but from his hands. Up close, through observation and trial and error, she internalized that tacit knowledge, which she later transformed into explicit, measurable movements for the machine. The corpus quote describes this process: “The definition of modeling offered… is the mapping of tacit knowledge into explicit knowledge.” [UNVERIFIED]

Vibe coding skips this step. You don’t go to the master baker. You type into the prompt: “Bake bread.” The AI generates something that looks like bread. But the rhythm of the kneading—that tacit layer ingrained in the body that distinguishes good bread from bad—is missing. You get the output, but you don’t understand the underlying why. You don’t see the movement of the hands, you don’t feel the resistance, you don’t internalize the process.

The SECI Spiral: A Blueprint for Knowledge Creation

The SECI model is not merely an academic framework; it is a functional description of how individual knowledge becomes organizational, sustainable knowledge. Let’s take a closer look at the four steps, which progress in a spiral, reinforcing one another:

1. Socialization (Tacit → Tacit): This is the first and most intimate form of transfer. No words are needed. The apprentice stands beside the master, observes, and imitates. In an open-plan office, sitting behind a senior developer, the apprentice sees how the senior navigates the debugger, what questions they ask themselves, and when they look up from the monitor. The corpus highlights that “socialization is typically triggered through human interaction and conversations” [UNVERIFIED]. This is the space of shared experience (“ba”), where knowledge becomes relatable.

2. Externalization (Tacit → Explicit): This is where the real challenge begins: transforming “knowing it by heart” into “this is how it’s done.” We try to articulate the tacit knowledge using metaphors, analogies, models, and sketches. A master baker might say, “It’s like stretching the dough just right.” A senior developer might draw an architectural sketch on the board or explain, “This function is like the conveyor belt in a packaging plant—it moves in one direction, and everyone knows who’s next.” According to the corpus, this process generates conceptual knowledge.

3. Combination (Explicit → Explicit): We organize, document, and connect pieces of explicit knowledge. Design documentation for a new system, code comments, corporate wiki pages. This is AI’s most natural domain: the systematic processing and recombination of information.

4. Internalization (Explicit → Tacit): Through practice, explicit knowledge becomes part of our body and mind. Just as a musician repeats scales until they become second nature, so too does a developer who performs many debugging sessions—the ability to sense “code smell” becomes ingrained in them. This is “learning by doing,” which results in operational knowledge and can serve as the starting point for a new cycle.

Where does the SECI spiral break down in vibe coding? The consequences of a technological shortcut

Vibe coding effectively begins at Step 3, Combination: explicit input (prompt) → explicit output (code). This radically shortens the path, but destroys the engine of knowledge creation.

• Step 1 (Socialization) is omitted: There is no shared space (“ba”) with the AI. There is no observation, no shared laughter over a bug, no thought process seen over our shoulders. AI does not share its intuition because it has none. The corpus mentions methods for transferring deep knowledge, such as “Guided observation” or “Socratic questioning” [UNVERIFIED]. These are all human interactions that a chat interface cannot replace.

• Step 2 (Externalization) is missing: Since the vibe coder lacks the initial tacit knowledge, there is nothing to externalize. It does not need to formulate the problem in its own words or search for metaphors. The prompt is an attempt at a quick, simplified externalization, but it is incomplete. According to Nonaka and Takeuchi, “the important part that constitutes most of knowledge creation is ‘the mobilization and conversion of tacit knowledge’” [UNVERIFIED]. This mobilization is missing.

• Step 4 (Internalization) is omitted: If you don’t write the code yourself, you don’t internalize its patterns, pitfalls, or elegance. Using generated code is like flying a plane on autopilot: it takes you from A to B, but you don’t learn to fly. Your hands never “get to know” the controls.

César Hidalgo writes in Why Information Grows: “Many jobs are largely based on intellectual capital or craftsmanship honed by years of education, training, and experience.” The promise of vibe coding: no need to hone your skills for years. AI just needs a prompt.

But honing your skills isn’t about learning syntax. It’s about making the tool an extension of your body. Just as a good carpenter doesn’t think about the weight of the hammer but “sees” the curve of the nail, a senior developer also sees the complex dependency between two modules before translating it into code. This is the tacit layer—and AI doesn’t generate it; it just bypasses it.

What is the difference between modeling and generation?

Here, it is useful to return to the concept of “modeling” mentioned in the corpus. In the context of Neuro-Linguistic Programming (NLP), modeling is “the mapping of tacit knowledge into explicit knowledge” [UNVERIFIED]. This is exactly what Ikuko Tanaka did with the master baker: she modeled his tacit knowledge to create an explicit algorithm from it.

Vibe coding is not modeling. It does not map tacit knowledge because it has no source. Generation. It combines and extrapolates from a massive corpus of explicit knowledge, but the spring of tacit knowledge does not flow into it from a tap. The difference is like that between a composer who notates the melody of a folk song (modeling) and an algorithm that generates a new song from all the Beatles’ songs (generation). One preserves an original, living layer of knowledge, while the other creates a probable variation of it.

How does vibe coding shape the developers of the future? The spiral of skill shortages

The question isn’t whether vibe coding works. It does—for small projects, prototypes, and one-off scripts. The real question is what kind of developer generation we’re raising if professional socialization is increasingly replaced by solitary prompt writing.

Imagine a junior developer who works mainly with LLMs at their first job. They successfully solve smaller tasks, but they never see how their senior colleague agonizes over a system design, how they sketch out dozens of possibilities on a piece of paper, and then cross them all out. They don’t participate in that painful, tacit knowledge-generating process of externalization. For them, the answer is ready, clear, and often functional. But the context of the decisions is missing.

Over the course of a generation, this cumulative layer of knowledge, this “professional intuition,” can fade away. Polányi’s tacit knowledge does not disappear—but it is not automatically passed on either. The figure in the corpus (Figure 2.6 Transferring deep knowledge) shows a whole spectrum of knowledge transfer methods, ranging from passive lectures to guided practice and experimentation [UNVERIFIED]. Vibe coding reinforces the most passive end of this spectrum: the delivery of a ready-made solution.

The baker’s hand “knows” when the dough is ready. The master programmer’s hand “knows” when the code smells. The vibe coder’s hand, however, merely types the prompt—and lacks the sensory faculties to understand what is missing.

Can the SECI cycle be revived in the AI era?

The challenge is not to reject AI, but to understand how to put the technology at the service of knowledge creation rather than bypassing it.

1. AI as an externalization partner: Imagine a senior developer beginning to explain a complex system pattern (an attempt to move from tacit to explicit). AI may be able to generate visualizations, analogies, or code snippets in real time that help make a thought—initially difficult to articulate—more precise. Here, AI does not replace socialization but rather reinforces the subsequent externalization phase.

2. Guided internalization with AI: We can view “guided practice” as the opposite of vibe coding. AI does not provide the complete solution, but rather assigns a series of progressively more difficult tasks while also providing guidance on how to solve them. The learner writes the code, but AI provides immediate feedback on style, efficiency, and alternative approaches. This is closer to “learning by doing” internalization.

3. Digital spaces for socialization: In the Nonaka model, the “ba” (shared space) could be physical. Today, it could be a virtual whiteboard, a pair programming session using VS Code Live Share, or even a recording of a senior team member thinking aloud about a problem. The goal here, too, is interaction and maintaining a shared focus.

The most dangerous path is when organizations view vibe coding as a tool for immediately boosting productivity while ignoring the disruption of the knowledge spiral. In the short term, more code is produced. In the long term, the organization’s “tacit capital” dries up, and there will be no deep knowledge to rely on for critical decisions (architecture, trade-offs, risk management).

Key Takeaways: What hands know, and the prompt does not

• Nonaka’s SECI model is not a theoretical framework, but a description of how knowledge is created: the transfer of tacit knowledge begins with socialization—vibe coding skips this driving force.
• The essence of the Matsushita bread maker example: Tanaka could only solve the problem through socialization (learning by observation) and externalization (formulating knowledge). The machine—and today, AI as well—only follows explicit instructions.
• Vibe coding starts at step 3 of SECI (Combination) — steps 1 (Socialization), 2 (Externalization), and 4 (Internalization) are skipped. You get working code, but you don’t build knowledge.
• Knowledge “grows” along a spiral: from the individual to the team to the organization. Vibe coding cuts off this “K-spiral,” as the corpus calls it, at a certain point [UNVERIFIED].
• A generation that has never “kneaded the dough,” never felt the resistance of the code between their fingers, will not know what is missing from the seemingly perfect generated code. The skill lies not in writing code, but in perceiving its absence.

Frequently Asked Questions

What is the SECI model?

The SECI model is Nonaka and Takeuchi’s knowledge creation framework, consisting of four spiraling steps: Socialization (tacit → tacit: knowledge transfer through observation and joint practice), Externalization (tacit → explicit: transforming tacit knowledge into metaphors and models), Combination (explicit → explicit: organizing and recombining known information), and Internalization (explicit → tacit: knowledge becomes second nature through practice).

How does AI affect tacit knowledge in coding?

AI, particularly in the form of vibe coding, bypasses the SECI cycle: it produces explicit output directly without the human going through socialization (learning from others) and internalization (acquisition through practice). As a result, the vibe coder may receive working code, but does not internalize problem-solving patterns, the context of decision-making, or the architectural considerations behind the code. The tacit layer of professional craftsmanship is not built up.

Can vibe coding be transformed into a knowledge-building tool?

Yes, but to do so, it must be consciously embedded into the SECI cycle. For example: AI does not provide the complete solution, but rather guidance and analogies (supporting externalization). The developer must actively review, modify, and understand the generated code (facilitating internalization). Most importantly, we must not abandon socialization—genuine human collaboration, pair programming, and mentoring—because this is the shared space where tacit knowledge can remain intact.

What does the concept of “ba” mean in the SECI model?

The Japanese term “ba” (場), introduced by Nonaka and Takeuchi, refers to the shared physical, virtual, or mental space in which knowledge is shared, created, and interpreted. This could be a physical meeting room, a joint project, an online forum, or even a shared set of values. The existence of such a shared space is a prerequisite for socialization. Vibe coding, or prompt writing done alone, often omits this shared space.

---

Related Thoughts

• The Polányi Paradox: Tacit Knowledge
• What the vibe coder doesn’t know that they don’t know
• The three types of tacit knowledge

---

Zoltán Varga - LinkedIn Neural • Knowledge Systems Architect | Enterprise RAG architect PKM • AI Ecosystems | Neural Awareness • Consciousness & Leadership The baker knows. The prompt doesn’t.

Strategic Synthesis

• Map the key risk assumptions before scaling further.
• Measure both speed and reliability so optimization does not degrade quality.
• Close the loop with one retrospective and one execution adjustment.