Skip to main content
LaRomierre
A person wearing headphones sits at a desk in a dark room lit by blue and orange neon lights, their hand on a large, illuminated "MERGE" key on a custom keyboard. They are looking at three monitors displaying code and text. A translucent, glitching red and cyan holographic projection of a man in a leather jacket and sunglasses, resembling a rocker, leans over their shoulder and points towards the central screen.

Not My Johnny Silverhand: How I Use AI Without Losing My Voice

• 6 min • 1,290 words
Artificial-Intelligence LLM-Workflow Cyberpunk-2077 Technical-Writing Code-Review Productivity AI-Ethics Human-in-the-Loop Developer-Experience

Lately, I’ve been analyzing my workflow. There is no denying it: AI has become a daily ally in everything I produce—whether in pure technical engineering, my role as a manager, or writing blog posts (which is the focus of this article).

After extensive experimentation, I want to share my feedback on what AI truly brings to the table, as well as the impact that automatic, abusive use can have on the quality of the message—risking trapping our output in an eternal cycle of rewriting what already exists.

In the Beginning, There Was Nothing…

… And from this nothing, AI was born.

AI barged into our lives with authority. For me, first contact goes back to my university days in computer science. Between two courses on computability and predicate logic—95% of which we assumed we’d never use—there was this Machine Learning module. We were told about neural networks and a form of artificial intelligence that, at the time, felt more like theoretical science fiction than tangible reality.

Then came the brutal acceleration of the late 2010s, driven by the Transformer architecture and the Attention Is All You Need paper. Fiction became reality, and “LLM” entered the common lexicon.

An Insidious Revolution

This isn’t humanity’s first technological revolution. But this one is more perfidious.

Unlike the internet or the smartphone, which you could choose to adopt or ignore, AI imposes itself on us through what it produces. It has two inseparable facets: the tool, which everyone is free to use, and the flow of generated content, which we all endure.

Today, platforms are flooded with content optimized for algorithms, produced at the lowest possible cost. It is a race for the “sugar rush”: that immediate reward humans insatiably seek with minimal effort. We have put a formidable tool in everyone’s hands, often transforming non-creators into generators of noise.

The result? AIs that are increasingly performant and realistic, yet placed in the service of content whose objective and veracity are becoming doubtful.

A group of four friends, two men and two women, laugh as they toast with glasses of beer in a warm bar with exposed brick and string lights. The woman in the center, wearing a leather jacket, has green reptilian eyes and scaly skin visible on her neck and hands as she holds her drink.
My (reptilian) friends and I at the bar

It is precisely to avoid participating in this ambient noise, and to guarantee that my use of AI serves meaning rather than volume, that I had to define strict guardrails.

Be My Player 2

In the beginning of my practical usage (early 2024), the AI wasn’t a production tool; it was a psychological safety net.

When I dove into the world of self-hosting, I was stepping outside my comfort zone. I wanted to set up dozens of services and launch a technical blog, but I lacked the specific engineering background to know where to start.

This is where the AI shone as “Player 2.” Unlike a senior colleague or a tech lead, the AI doesn’t judge. I could ask “naive” questions, expose my lack of knowledge, and debate implementation details at 2 AM without fear of looking incompetent. It gave me the courage to attack subjects that previously felt inaccessible.

The Trap of the “Codebase” However, I quickly hit a wall. The AI would often provide entire codebases or configuration files to implement. At first, I fell into the trap of mechanically copy-pasting its output. The result? It rarely worked. I found myself in endless loops of copy-pasting errors back into the prompt, asking for fixes, and getting new broken code in return.

I had to learn to “look up.” I realized the AI is an icebreaker, not an oracle. It provides an angle of attack, but cross-referencing its suggestions with official documentation is the only way to actually make things work.

Be My Senior Reviewer

Once the honeymoon phase was over, I had to professionalize the relationship. To produce high-quality work, I don’t need a cheerleader. I need friction.

My current workflow is designed to kill complacency:

1. The “Mac and Cheese” Test By default, an LLM is a “Yes-man.” If you ask it to rate your terrible mac and cheese recipe, it will give you a 9/10 and tell you it’s “inspiring.” I don’t want flattery; I want objectivity. I explicitly instruct the AI to challenge my premises. If I pitch a generic idea that has been written a thousand times, I want it to tell me to scrap it.

2. Structure over Substance I provide the chaos; the AI provides the order.

  • Ideation: I feed it my raw, unstructured stream of consciousness.

  • Literature Review: I use it to scan existing content to ensure I’m not just paraphrasing what’s already out there.

  • Refinement: Since English isn’t my first language, the AI acts as a syntax and semantic filter.

3. The Golden Rule: No Fillers The AI is allowed to cut fat, rephrase clunky sentences, or suggest a better flow. But it never generates new content. If it thinks a section is missing, it must justify why. If I agree with the logic, I write the paragraph. The AI processes; it does not create.

Not My Johnny Silverhand

This is the most critical part of my philosophy. In Cyberpunk 2077, Johnny Silverhand is a charismatic construct in your head who slowly overwrites your neural pathways until you cease to exist.

AI has the same potential. It has a “statistical voice”—smooth, confident, and utterly generic. If you aren’t careful, your unique tone gets overwritten by the algorithm’s average.

To prevent this “Cyberpsychosis,” I treat every AI interaction like a Pull Request (PR).

In software development, dropping a ‘LGTM’ (Looks Good To Me) without reading the code is a **professional sin• . The same applies here. When the AI suggests changes, fixes typos, or refactors a paragraph, I treat it exactly like an incoming Pull Request to my repository.

The LLM acts as a reviewer, auditing my work based on the guidance I provided in the preamble. It is then up to me to review these suggestions against the story I want to tell, my communication style, and my lived experience. I accept, decline, or adjust the content based on that feedback—whether it’s a structural rewrite or simple grammar fixes to iron out my language mistakes.

The Bottom Line Like a Lead Developer reviewing code, the AI advises and proposes. But I hold the Merge button. I never let the AI “auto-merge” its suggestions. The moment you stop reviewing the output is the moment the article is no longer yours—you’re just the host body for the machine.

A Final Retrospective

This strict workflow is not just a personal preference; it is a survival strategy for authorship.

We have reached a tipping point where most of the content is now generated via or using AI-powered tools (see Graphite.io report). This saturation extends beyond blog posts; it is even impacting the rigorous world of scientific publications (see ScienceDirect study). In this ocean of synthetic text, preserving your human intent is the only way to stay relevant.

As I wrap up this post, I verified that the writing process for this very article followed the exact methodology described above to ensure it didn’t drown in that noise:

  • The Input: The core concepts—the self-hosting anecdotes, the Pull Request analogy, and the “Silverhand” metaphor—came strictly from my experience.

  • The Review: The AI acted as a sparring partner. It cleaned up my English and challenged the structure, but it did not generate the ideas.

  • The Merge: I rejected its attempts to force pop culture references (like Iron Man’s Jarvis) or to heavily layer the Cyberpunk metaphor (suggesting slang like “choom” or “cyberpsycho”), which would have undermined the article’s credibility.

This article is a practical application of the workflow. The AI helped refine it, but the “Merge” button remained under my control.

Further Reading & Inspirations