On not letting an LLM write this.

I am writing this essay, and not letting a language model write it, and the reasons matter more than the act.

The temptation to use one was real. Roughly two years into building Wouch, the team had shipped most of the platform; the marketing site needed regular updates; the Signals masthead was about to launch and would need content at a publication cadence we did not have the founder hours to sustain. We had access to capable models. A team meeting in early 2025 surfaced the suggestion: a careful prompt, a strong editorial layer, and we could produce something nominally indistinguishable from human writing in a fraction of the time.

The case for using one is genuinely good. Speed. Volume. Voice consistency, in theory, if the prompt engineering is tight. The argument that the founder's time is better spent on platform than on writing essays. The observation that competitors are increasingly doing exactly this and that resisting it is an obstacle, not a choice.

I have thought about each of these for a long time, separately and together. The conclusion has held in every version of the analysis: we will not use a language model to write what a user reads here.

There are three reasons. The first is about you, the reader of this essay. The second is about me, the person writing it. The third is about Wouch as an institution.

About you, first. When you read something on the Signals masthead, you are reading it for a reason that does not survive automation. You came here because something about Wouch's positioning landed with you - the refusal to use urgency, the language that treats you as a thinking adult, the specific kind of restraint that distinguishes the platform from its commercial neighbours.

An essay that you later discovered to be model-generated would retroactively break that contract - even if the prose was good, even if you could not have told at the time. The trust is in the human-on-the-other-end. Removing the human, however gracefully, removes the trust.

About me, second. The act of writing this essay is part of how Wouch stays itself. There is no shortage of internal documents at Wouch - the operational corpus runs to seventy-some documents now, each careful, each reviewed. Those documents are not enough on their own to hold the substrate.

The broader claim underneath these three reasons is the one I want to leave you with. Every platform that uses language models for its substantive content is implicitly making a choice about what kind of relationship it has with its readers. The choice is mostly invisible because the prose is good and the readers cannot tell.

I am aware that this position will get harder. The models will get better. The competitive disadvantage will increase. A version of this same essay will need to be written then, addressing whichever new argument has emerged for crossing the line. I expect to write it.

Thank you for reading what was written.