INDEPENDENT JOURNAL / AI + ANALOGEST. 1988
Back to the journal
  • GenAI
  • Human input
  • Field notes

I did not build this post; GenAI did it for me

GenAI wrote the words and built the site. Here is the human input behind it: the brief, the context, the decisions, the corrections, and the responsibility.

GenAI drafted this post. It also wrote the code for this website and handled the technical work of getting it online.

I didn’t type every sentence or write the components. I supplied the intention, the context, and the decisions that made the output mine to stand behind.

This first post is a record of that contribution, from the initial brief to the version you’re reading.

I gave it a reason to exist

The starting point was a simple request: build a blog-like website for Nespera Cosmica, make it brutalist, host it on my existing DigitalOcean server, and connect the domain I had bought.

That was enough to start a conversation. It wasn’t enough to define the result.

When the proposed direction leaned towards a company introduction, I clarified the purpose: share knowledge through a feed of posts. Contact should lead to my LinkedIn. Merchandise should have its own section.

That correction changed the site’s priorities. The writing became the main event.

I made the small decisions that shape the whole thing

English. Excerpts in the feed. Full posts on their own pages. Markdown files for publishing. A merchandise page marked “Coming soon”. An existing server rather than a new hosting account.

None of those decisions writes a website by itself. Together, they define what the system should build, what it should leave out, and how I’ll maintain it later.

The questions were part of the work. Answering them gave the model constraints it couldn’t reliably infer from “a simple blog”.

I supplied context the model didn’t have

I brought the name, the domain, the server choice, the logo, and the LinkedIn destination. I signed into the domain provider so the setup could continue. I later supplied the SVG version of the artwork.

The model could inspect the environment and carry out the technical steps. It couldn’t invent my ownership of those things or decide which of my resources I intended to use.

Useful input includes assets and access, as well as words.

I told it the first version wasn’t there yet

The first site worked. My response was: not brutalist enough.

I gave it stronger references: Intro to Brutalism in Web Design and DOOM. I supplied the message I wanted at the centre of the site:

Work in AI; Rebel Analogically

I asked for a logo back print on an oversized T-shirt. I changed the masthead to EST. 1988. I supplied a sketch for the small line mark in the header. And I replaced the generic welcome post with the title you clicked to get here.

Those inputs are art direction and editorial judgment. The model can produce a plausible interpretation of “brutalist”. It needs my feedback to learn whether that interpretation is the one I actually want.

I asked for a way back

I also asked for Git version control so I could restore an earlier version.

That matters when the system can change a lot of things quickly. A saved baseline gives me something concrete to compare with the next iteration. Server releases provide another way to restore the deployed site.

The model ran the build and deployment checks. Asking for a recoverable process was my contribution; implementing and checking that process was part of its work.

The input continues after the first prompt

The sequence here was straightforward:

  • Set a purpose and an initial brief.
  • Answer questions about audience, content, publishing, and hosting.
  • Supply the real assets and destinations.
  • Review the result and identify what felt wrong.
  • Give more precise references, language, and corrections.
  • Keep a version history and decide what to publish.

The generated output was material for the next decision. “Not brutalist enough” was input too.

I still own the judgment

The model generated this explanation of my contribution. That is worth stating plainly.

I’m responsible for the decision to publish it, for checking that it represents what happened, and for correcting anything it gets wrong. Generated words don’t acquire accuracy just because they sound confident.

Working in AI leaves plenty of room for human taste, disagreement, and life away from the screen. That’s what the message on this site means to me.

Work in AI; Rebel Analogically.


Production note: this article was drafted by GenAI from the conversation used to create and revise this website. The T-shirt image is a generated concept mockup, not a photograph of a manufactured garment.