Redaktor: writing between human craft and artificial technique
Something is changing in the way human beings produce text. We are witnessing a qualitative transformation that simultaneously alters the process, the product, and the author’s relationship with their own work. Generative artificial intelligence has ceased to be an assistant that fixes errors or suggests synonyms: it is an agent capable of composing, structuring, evaluating, and rewriting with a degree of competence that forces us to rethink what it means, today, to write.
This essay narrates a concrete experience of that transformation. I write in first person, from the pursuit of improving my writing skills.
The starting point: a conversation with the machine
It all began with a simple question posed to ChatGPT: how can I write better in Spanish? What came back was an iterative conversation that helped me identify my weaknesses: imprecise punctuation, excessive subordination, vague adjective use, scarce rhythmic variety. At that stage, the machine served as a diagnosis: it pointed out problems with the candor of someone who has no ego to defend and no reputation to protect.
That conversation led me, a few months ago, to the Curso de redacción by Gonzalo MartÃn Vivaldi, a systematic treatise spanning from punctuation to literary composition techniques. I decided to study it with the rigor of someone training in a manual craft: chapter by chapter, taking notes, testing rules, working through the exercises. During that period, the learning was entirely human. Reading, reflection, practice, error. No AI.
Vivaldi’s book turned out to be an extraordinary work. Across more than five hundred pages, it condensed a complete system of principles, operational rules, and examples covering everything from the vocative comma to the technique of free indirect style. The real difficulty lay in integration: writing a paragraph that simultaneously respected conciseness, naturalness, lexical precision, verbal dynamism, and sonic harmony demanded a capacity for multifocal attention that can only be acquired through practice.
The need to systematize
After finishing the course, I faced a practical problem. The knowledge was in my head, scattered among notes and highlights, but I lacked a mechanism to activate it operationally every time I sat down to draft a new text. What I needed was a system capable of translating that mass of principles into concrete, weighted, and adaptable instructions tailored to the genre, audience, and purpose of each piece of writing. Something more ambitious than a summary, more structured than a set of notes.
Redaktor: a professional writing engine
Redaktor is a web application that converts the knowledge from the Curso de redacción into an operating system for AI-assisted writing. Its architecture is organized around three concepts: a knowledge base extracted from the book (principles, rules, anti-rules, transformations), an ontology of stylistic variables (lexical precision, conciseness, naturalness, verbal dynamism, use of metaphor, among others), and a set of writing profiles that assign differentiated weights to each variable depending on the type of text to be produced.
The application itself dispenses with any language model. Its function is to generate prompts: structured, complete instructions that the user copies and runs in the model of their choice. These prompts incorporate the writer’s role, the relevant knowledge base, the active profile with its weights, the specific task, and the execution instructions. The result is a text that responds to explicit and measurable quality criteria — far removed from the generic intuition of a model operating blindly.
In addition to the generation module, Redaktor includes an evaluation module that produces structured critique prompts: overall score, per-variable analysis, strengths, detected problems, improvement suggestions, and a rewrite of the weakest paragraph. An import module allows the knowledge base to be fed with new chapters processed by an external model, and a visual graph displays the reinforcement, tension, and dependency relationships among the system’s variables.
The process: vibe coding and agentic recursion
The implementation of Redaktor followed a method commonly known as vibe coding: AI-guided programming where the developer defines the intent and the agent generates the code. I used an agentic workflow — an AI model with access to reading, writing, and code execution tools — that allowed me to have a functional application within a couple of hours. The tech stack (React, TypeScript, Vite, Tailwind CSS, Zustand for state management, React Flow for the graph) was selected by the agent based on an implementation plan that it generated itself.
Here an interesting trait emerges: that plan was produced thanks to prompts I had previously built for similar software design tasks, and those prompts, in turn, had been refined with the assistance of artificial intelligence. The process reveals its recursive nature: each layer of automation builds on a previous one, and each iteration produces a result constructed from the input of its predecessor.
One more detail worth adding: this very essay was generated by artificial intelligence and only editorially revised afterward by me. The prompt that produced it was built, precisely, with Redaktor.
Writing, enframed
The process I am describing is not anecdotal — it is an instance of a structural transformation affecting every creative process mediated by language. Artificial intelligence, at least for now, still requires a writer on the other side — but it redefines their role: it shifts them from executor to architect, curator, and editor.
Martin Heidegger warned, in his reflection on the essence of technology, that the technological far exceeds the instrumental. What he called Gestell — enframing, the structure of provocation that converts every being into available resource — designates a mode of unconcealment that configures our entire relationship with reality, beyond any individual will to control. Generative artificial intelligence operates precisely this way: it constitutes a framework that reorganizes what we understand by writing, authorship, and thought. When a model can generate a technically correct essay in seconds, the pertinent question shifts from how to write to what it means for someone to write.
Marshall McLuhan complements this perspective with an equally incisive insight: every medium is, simultaneously, an extension and an amputation. The wheel extends the foot and atrophies the capacity to walk long distances; writing extends memory and weakens oral tradition. Artificial intelligence extends the writer’s compositional capacity — enabling them to produce more, faster, with greater control over stylistic variables — and, at the same time, exacts a price: perhaps the reflective slowness that compels one to weigh every word, perhaps the resistance of the material that strengthens the craft, perhaps a certain intimate relationship between expressive effort and the authenticity of the result.
Conclusion
Redaktor embodies this tension without pretending, in any way, to resolve it. It is, after all, a weekend project to entertain my restless and creative spirit. What I find interesting is showing how the creative process has been transformed by the irruption of AI and how it allows us to bring to life ideas that, in the past, would have been impossible to materialize within the schedules and responsibilities of our daily lives. Every day I come across apps created by people who have never programmed in their lives, and it fills me with interest and curiosity to see ideas from people who were never in tech brought to reality. I hope AI serves to foster the creativity and spirit of many people, especially those who were never part of the technology sector.