Beyond Vibe Coding: A Designer’s Case for Directed Generation

The term "vibe coding," coined in early 2025 by AI researcher Andrej Karpathy, initially served as a precise descriptor for a specific mode of interacting with generative AI models: loosely defining an objective, accepting the output with minimal scrutiny, and exhibiting low accountability for the results. However, this term has migrated from its accurate, albeit limited, context into broader discussions surrounding AI-assisted design, a semantic shift that designers argue is actively undermining the authorship and perceived value of their practice. This evolution of language is not merely a semantic quibble; it carries significant implications for how design work is understood, commissioned, and ultimately, valued in an increasingly AI-integrated professional landscape.

The core of the issue lies in the appropriation of "vibe coding" to describe more intentional and sophisticated design methodologies. When designers fail to challenge this label, they risk relinquishing control over the narrative of their own expertise. The true divergence in AI-assisted design is not about adoption, but about authority – specifically, where the locus of judgment resides within the creative process. In the realm of "vibe coding," the generative model’s output dictates the direction, with human input acting as a superficial layer of adjustment. Conversely, a more advanced approach places human judgment firmly at the forefront, guiding the AI’s response with deliberate intent.

The Genesis and Migration of "Vibe Coding"

Andrej Karpathy’s original articulation of "vibe coding" was intended to capture a particular, low-stakes interaction with AI. It described a scenario where a user might offer a vague description, accept the generated output without deep analysis, and move on. The term’s accuracy stemmed from its specificity to this passive and less accountable approach. Karpathy himself acknowledged this in early 2025, framing it as a descriptive label for a observed behavior.

The problem arose when this precise term began to transcend its original confines. As is common with nascent jargon, "vibe coding" proved to be a readily adoptable shorthand, even when misapplied. Designers have reported their own rigorous workflows being dismissed as "vibe coding," and many have passively accepted the label without pushback. This uncritical adoption allows a superficial understanding of AI-assisted design to gain traction, obscuring the nuanced judgment, intricate craft, and deliberate intentionality that characterize serious design work.

The power of language, as George Orwell famously observed, extends beyond mere communication; it shapes thought itself. Imprecise terminology can foster imprecise thinking, and when an ill-fitting label colonizes an emerging field, it preemptively defines the practice before its true contours can be established. This can lead to misunderstandings in hiring decisions, project valuations, and the overall perception of a designer’s capabilities. The current vocabulary surrounding AI-assisted design is still in its formative stages, with much of it originating from engineering culture. This influx carries inherent assumptions, often positioning the human role as supervisory at best, and fundamentally passive at worst.

The Emerging Divide: Authority and Intentionality

A critical division is solidifying within the field of AI-assisted design, a split that transcends mere adoption rates. While nearly all designers are now experimenting with AI tools, the fundamental difference lies in the locus of authority within their workflows.

One prevalent approach, often inadvertently labeled "vibe coding," prioritizes generation as the primary driver of decisions. This process typically involves:

• Prompting: Initiating the AI with a request.
• Acceptance: Taking the initial generated output with minimal critical evaluation.
• Marginal Adjustment: Making minor tweaks to the existing output.
• Shipping: Releasing the work with limited deep-dive analysis.

This model, while efficient for certain tasks, places the AI’s generative capacity at the center, with human input playing a reactive rather than proactive role.

In stark contrast, a more advanced and increasingly prevalent methodology places human judgment as the initial and guiding force. This "directed generation" approach operates on a fundamentally different premise:

• Judgment First: The designer’s expertise, informed by strategic goals, aesthetic principles, and user needs, dictates the direction.
• Curated Input: Instead of a broad prompt, designers utilize specific, curated references – sketches, mood boards, visual precedents, or even existing design elements. These references are not mere suggestions but carry embedded decisions about proportion, tone, hierarchy, and intent, often far more precisely than verbal articulation.
• AI as a Responding Agent: The AI is tasked with responding to this established human intent, acting as a sophisticated tool for iteration and exploration rather than a primary source of creative direction. The designer guides the AI, not the other way around.

This more deliberate approach is harder to encapsulate with a single, catchy term, which has allowed "vibe coding" to fill the linguistic vacuum. However, it is this method that is enabling serious design work to be executed at unprecedented speed and scale, without sacrificing quality or intentionality.

The Foundation of Directed Generation: Intentional References

The practice of directed generation begins not with an open-ended prompt, but with a deeply considered reference. This reference is the embodiment of the designer’s judgment, carrying pre-existing decisions about critical design elements. A designer might present a meticulously crafted sketch, a curated screenshot, or a collection of visual precedents. These inputs are rich with encoded information regarding:

• Proportion and Scale: The relative size and relationships between elements.
• Tone and Mood: The emotional resonance and stylistic character of the design.
• Hierarchy: The visual ordering of information and emphasis.
• Intent: The underlying purpose and message the design aims to convey.

These references function as a highly compressed form of design specification, far more precise and evocative than lengthy textual descriptions. Designers have historically relied on such visual communication, but the advent of AI tools capable of interpreting visual intent at the fidelity of code has revolutionized this process. The AI becomes an extension of the designer’s thought process, capable of rapidly translating abstract visual concepts into tangible outputs.

This process then unfolds through distinct phases, and the mastery of knowing when to transition between them is the hallmark of skilled practitioners. The infographic below illustrates a conceptual framework for these phases, highlighting the iterative and guided nature of directed generation.

([Infographic on directed generation phases would be inserted here if visual content were possible]) – The infographic, created by Jim Gulsen, visually represents the multi-stage process of directed generation, emphasizing the designer’s active role in guiding the AI through various creative and refinement stages. It likely depicts phases such as reference input, initial generation, iterative refinement, and final output, underscoring the human-driven decision-making at each step.

The Transformative Outcome: Scalable Design Without Stagnation

The implications of embracing directed generation extend far beyond individual project execution. They fundamentally alter how design systems are conceived and how design can scale without becoming rigid and inflexible.

In traditional design systems, patterns are often treated as fixed artifacts: defined, meticulously documented, and applied rigidly. Consistency is achieved through replication, but this often comes at the cost of flexibility. When faced with new contexts or unforeseen requirements, these systems can become a constraint, hindering innovation and adaptation.

Directed generation, however, liberates design systems from these constraints. When a machine can grasp a design pattern at an abstract level, it gains the capacity to recompose it contextually, rather than merely copying it. This means a pattern is not just a static component but a set of underlying spatial, typographic, and behavioral relationships. The AI can then interpret and apply these relationships across novel surfaces, devices, and contexts it has never encountered before.

The designer’s role thus evolves from specifying every minute detail of each instance to defining the conditions under which high-quality, coherent instances will reliably emerge. This shifts the focus upstream, to the quality of the foundational elements, the precision of the references provided, and the rigor of the constraints established.

This leads to what can be termed "non-deterministic design." The output is not entirely predictable, but this unpredictability is not a flaw; it is an inherent characteristic of a system designed for adaptability and emergent creativity. The craft is amplified in the upstream definition of the design grammar, enabling the system to speak it fluently.

The natural endpoint of this evolution is "agentic delivery." This refers to systems that not only respond to design direction but can autonomously carry it forward, generating coherent outputs across vast scales and diverse contexts without sacrificing the authoring intent embedded in the original primitives. This represents systematic design generation, moving beyond mere systematic design documentation. The designer authors the rules of the language, and the AI becomes a fluent speaker capable of articulating them in myriad ways.

This trajectory is not about finding workarounds for AI limitations; it is about defining the future capabilities of AI tools by grounding them in robust design principles and intentional human guidance.

Navigating the Lexicon: Towards Precise Terminology

The process of naming an emerging practice is inherently challenging. A term that is too premature can calcify before the practice has fully matured, while a term that arrives too late risks being overshadowed by established jargon. AI-assisted design is currently in a phase where its methodologies are solidifying, but its vocabulary is still heavily influenced by external disciplines. Terms often originate from engineering, product management, or the sensationalized narratives surrounding generative AI, where every output is treated as the ultimate goal rather than a means to an end.

To foster a more accurate understanding and appreciation of this evolving practice, precise terminology is crucial. While not intended as a rigid taxonomy, several terms offer more descriptive handles for the work being done:

• Directed Generation: Emphasizes the human-led, intentional nature of the AI interaction.
• Reference-Guided Practice: Highlights the foundational role of curated inputs in shaping the AI’s output.
• Compositional Prompting: Focuses on the AI’s ability to compose and recompose elements based on designer-defined structures.
• Intentional AI Collaboration: Underscores the partnership between human intent and AI capabilities.
• Pattern-Based Generative Design: Stresses the understanding and manipulation of underlying design principles.

These suggestions are not mandates but rather starting points for a more precise discourse. The ultimate goal is not to win a naming debate, but to equip designers with language that enables clearer thinking and more effective communication about the value of their work. As the article argues, those who actively define this practice will possess greater authority than those who passively inherit someone else’s simplified labels.

(Infographic on naming directed generation practice would be inserted here if visual content were possible) – This infographic, also by Jim Gulsen, likely explores various proposed terms for the practice of directed generation. It might offer visual metaphors or brief definitions to help differentiate these terms and highlight their strengths in capturing the essence of intentional AI-assisted design.

The Pragmatic Imperative: Why Naming Matters

A common and understandable objection to this discussion is: "Does it really matter what we call it? The work gets done either way." Clients often focus on the final output, and designers are inherently pragmatic individuals focused on delivering results.

However, the significance of precise language becomes critically apparent in specific, often underestimated, professional contexts:

• Portfolio Reviews: When explaining one’s methodology to potential employers or collaborators, clear articulation of process is paramount.
• Client Kickoffs: Establishing the unique value proposition of a designer’s approach, particularly when justifying rates, hinges on the ability to describe a sophisticated and effective process.
• Team Conversations: Junior designers seeking to understand what constitutes "craft" in this new technological landscape require models and language that accurately reflect intentional design practice.

In these crucial moments, possessing precise, confident, and personally defined language for one’s work translates directly into professional clarity and credibility.

This situation mirrors the emergence of User Experience (UX) design as a distinct discipline. The practitioners who actively named and defined their work—identifying themselves as interaction designers or information architects—shaped how the field was valued for a generation. Those who allowed adjacent disciplines to define them spent years trying to reclaim ground and establish their unique identity.

AI-assisted design is at a similar inflection point. While the technology itself is no longer novel, the practice is still being codified. The definitions that will endure will either emerge from engineers describing their shipped products, journalists covering the most visible outputs, or designers articulating the actual intricacies of their craft.

The traditional, deterministic model of design (producing fixed components, documented patterns, and predictable handoffs) remains essential for many production environments. However, the value of directed generation and non-deterministic design lies not in replacing these established methods but in expanding the capabilities of design systems. Areas like generative wireframing, rapid concept exploration, and cross-surface pattern interpretation yield outsized returns through non-deterministic approaches.

A hybrid approach, acknowledging that different phases of design work carry varying risk tolerances, is often the most mature strategy. Production-ready components may still require specification and rigorous review, but the early stages of ideation and exploration benefit immensely from the flexibility and speed offered by non-deterministic methods.

"Vibe coding" accurately describes one way of interacting with AI, and for certain contexts, its low-accountability nature might be appropriate. However, it has never been a fitting descriptor for design leadership. The sooner this distinction is universally understood, the sooner AI-assisted design work will receive the recognition and framing it genuinely deserves. Whether termed "directed generation," "reference-guided practice," or "compositional prompting," the adoption of precise language empowers designers to own their practice and, consequently, to shape its future. The choice is to inherit a simplified label or to define the work with the authority it commands.