How I write essays in 2026

I’ve updated my writing process and workflow for 2026, and the main reason I am revisiting it is because the process itself has started to change in a meaningful way.

Writing is no longer just about producing an essay. It has become a feedback loop where each draft I produce also acts as training data. I’ve naturally been interested in the gap between what AI generates and how I would naturally write is something I am now actively trying to close. Every piece I write contributes to that. The process I am describing here is not just how I write, but how I am shaping future AI outputs to sound more like me. In a way, this essay itself is part of that loop, and I’m naturally interested in the end game, where the AI sounds exactly like me; we’re not even 50% there yet, but curious how this would unfold..

While doing this, I’ve also ensured the principle source of the “key idea” still comes from me, it should resonate with me first, and the integrity of this nature continues to remain uncompromised.

To talk about my writing process, it begins long before I open a document. It starts with a running surface of thoughts that I maintain in Trello.

This is where everything goes first, without pressure to make sense. Ideas arrive as fragments, questions, contradictions, or half-formed intuitions. I don’t try to structure them immediately. What matters is that they are captured quickly. Over time, some of these ideas begin to stand out. They survive small acts of revisiting, get slightly reworded, or start to point in a clearer direction. Sometimes I had a good image thumbnail, to entice myself in drafting the essay further. When an idea feels like it has “hit home”, it shapes up to be explored further, it stops being just a card and becomes something I want to work on deliberately.

That is the point where I move it into Obsidian. This is where the rest of my personal knowledge base resides, and which I now try to take advantage of through a compounding loop which I will talk later about.

I create an Obsidian note with the intention to shape this into an essay, YouTube video, Substack article or anything else. Around it, I begin to gather material. This includes articles, essays, research, and most importantly, my own past writing that might connect to the same theme. The vault becomes a kind of extended memory. I follow links, revisit old notes, and slowly build a context around the central idea. Sometimes this expands what I already think. Other times it challenges it. Either way, the idea becomes less isolated and more dimensional.

As the material builds up, I move into a more deliberate way of processing it using the Zettelkasten approach. I have started to think of this stage as a process of crystallization. Ideas, when they first appear, are in a gaseous state (in this case, the ones that are first documented on Trello). They are diffuse, free-floating, and difficult to pin down. You can sense them, but you cannot yet hold them. As I begin working through notes, something changes. Writing reference notes and especially literature notes from memory forces the idea to condense. It becomes more coherent, more bounded. This is the liquid state. It can still flow and change shape, but it has a form that can be worked with.

During this process of working with my drafts on Obsidian, I also use tools like Enzyme, to start making more semantic connections with my existing notes on the vault. Connections appear between ideas that did not seem related at first. And then, in writing permanent notes, the idea crystallizes. It becomes solid. It takes the form of a single, clear, self-contained thought expressed in my own words. At this point, it is no longer something I am trying to understand. It is something I can use.

The Zettelkasten process, for me, is fundamentally about taking ideas through these states, from gas to liquid to solid, until they become stable enough to build with.

Some ideas still feel unclear even after this. They might have internal tension or multiple possible interpretations. Or might even lack proper framing.

In those cases, I use the Hegelian dialectic to shape them further. I take an initial framing and push it as far as it can go. Then I construct an opposing view and strengthen that as well. The goal is not to balance the two, but to force both sides to become sharp. What emerges from this is usually a better articulation of the idea. Sometimes it is a clearer position. Sometimes it is a reframing that resolves the tension altogether. Often, it simply results in better wording that feels more precise.

Once the idea feels clear enough, I start assembling it into a draft. This is where I bring together the different notes and arrange them into a sequence that makes sense. I pay a lot of attention to what deserves to be central and what should remain in the background. Not every idea should carry the same weight. There is usually one core message that I want to drive, and everything else either supports it or stays peripheral. This shaping is important because without it, the writing tends to become scattered, with too many ideas competing for attention.

If I have used any AI-generated text in the process, I clean it up at this stage. I remove phrases that feel generic, sections that over-explain, and structures that feel predictable. The amount of AI-generated text might vary essay by essay. Sometimes I just ramble on my phone (ChatGPT dictation on chat threads work quite well), and then convert this messy rambling into a AI generated first draft.

The goal is not just to simplify, but to make the writing feel specific and grounded. After this, I don’t treat the draft as final. Instead, I use it as a reference and start writing again from scratch. I type everything in my own words. This helps me rebuild the flow in a way that feels natural to me. While doing this, I also pay attention to how I feel about what I am writing. There are moments where the writing needs more intensity and others where it needs restraint. This emotional alignment is something I can only achieve when I am actively rewriting rather than editing.

At the end of this, I have two versions of the same piece: one that came out of the structured process and another that reflects how I naturally write. I treat this as an input-output pair to train a custom agent skill that I’ve developed to make the AI generated writing write like me.

Over time, I am collecting these pairs with the intention of using them to train models that can better match my style. The idea is that the gap between what the AI produces and what I would write myself keeps shrinking. It is not something that happens instantly. It improves gradually as more examples accumulate, and every essay I write contributes to that convergence.

Once the writing is ready, I turn it into an artefact. Sometimes it remains a blog post. Other times it becomes a rough note or evolves into something larger like a video.

Alongside this, I am building toward a system where a persistent agent takes over distribution. I use a Hermes agent for this, a self-evolving memory layer that can run independently and learn from how I write, what I publish, and where it performs best. The direction is for it to eventually handle publishing across platforms on my behalf, adapting the same core idea to different contexts without losing the original intent. It is still a work in progress, but it is the natural extension of the system I am building.