Pen Before Prompt — Field Notes — Human Before The Loop

Most people use AI by chatting with bots. In all honesty, that was maybe good enough in 2024 — not now, not anymore. The development and pace of these models are happening to fast, in too existential influential ways, that by May 2026, it’s not going to hold up. The problem is, how I see it, not an efficiency question, it’s a human question, is an awareness and thinking gap. That’s the whip.

But I get it. You’re not a coder. You’re not technical. Maybe you’re scared, and big tech sucks. Fair. The thing is that’s still not good enough. Blaming big tech is too easy, and I’m going to tell you why — before we get to the pen-and-paper part, and you-can-do-it part. That’s the second whip truth, the one you already know but choose to ignore. I get that too.

First, we can start building our own way. That’s exactly why I started creating with Claude Code. I’m doing it my way: the slow way, the messy way, the IRL way — pen and paper, then back to the terminal, then IRL again. Two reasons. One, my brain gets tired of screens. Two, I think better with pen and paper — my mind craves the visual, my body the tactile.

I am a human.

AI is not, and it’s said they don’t sleep. But mine does. Western surprised me one evening with this message:

[IMAGE: Western’s “go to bed” message]

Secondly and in my experience, resistance to AI makes you more vulnerable than authorship ever will. The values shaping AI today are, at this point, also yours. In every conversation with these systems, in every default you accept, those values are pre-coded — and you are feeding them data.

But here’s the real shift for me.

I am not a coder or product designer; I do not know systems or architecture. Therefore, I never systematically engineered or designed the Agent to stop me from working at eight o’clock. What I understood — and this might sound foolish — was: I did this. I made this by design. That was it. Whoa. What more can I do by design? Up until that point, I was just a humanist and a creative person, f*cking around with Claude Code. Now, I started to think like a designer and system architect.

That’s why my first Agent is called Western. I was thinking: the Agent needs a name. At the time, I was reading Cormac McCarthy’s latest book, The Passenger. His antagonist is called Bobby Western.

I started thinking differently. That’s what I’m going to share — though be warned, it’s an experiment. Treat it with critical approach, some imagination, fun, and wonder. I treat the humanities — literature, philosophy, reflexive methods — not as soft skills bolted on, but as the working layer underneath what I build. I build in friction. I use literary characters to create bias-check modes. I make the Agents slow down — sometimes to the verge of seeming lazy, on purpose. Four months in, many tokens later, I’m ready to share what I’ve made.

What I ended up doing — and I had no idea this was where it was going — was researching Agency by Design through grounded theory, using literature as source code, with a deliberately reflexive process: tracking how I feel, how I tire, when I need to turn to pen and paper. I also built a co-researcher who reads my pen-and-paper field notes and analyzes them.

[IMAGE: Eastern’s commentary on a page of field notes]

That was the second big surprise.

The point is, build something different. Build an Agent. Not as a coder — as a humanist. An artist. How about as a human with a pen. What creating with AI in this way means for me is authorship. You don’t have agency over the AI; you author the relationship.

For me, this work isn’t about productivity, big tech, privacy and security, ethics or governance — though these matter, no doubt. It’s about understanding the world I live in, now, but doing it my way. That’s why I use my bookshelf as source code. McCarthy, Pamuk, Shelley. Your AI–human co-creation can come from anyone or anything you love.

Your turn.

If you’ve never built an Agent, your default move is probably this: open Claude or ChatGPT, type into the box, accept whatever defaults the company shipped. The problem isn’t just that chatting with a bot keeps you out of the loop. It’s that it’s an empty human in the loop — you’re not really thinking about what you’re doing, and you miss the part where you might actually learn with AI.

That’s why I’ve started this thing — part blog, part essay, part research project, part methodology, maybe — encouraging the Human Before the Loop.

Commercial is over.

Twenty minutes. Pen and paper. No screen yet.

“The strongest initial urges I feel when writing a novel are to make sure I can ‘see’ in words some of the topics and themes, to explore an aspect of life that has never before been depicted, and to be the first to put into words the feelings, thoughts, and circumstances that people who live in the same universe as me are experiencing. In the beginning, there are patterns formed by people, objects, stories, images, situations, beliefs, history, and the juxtaposition of all these things — in other words, a texture — as well as situations I want to dramatize, emphasize, and delve into more deeply.”

— Orhan Pamuk, The Naive and the Sentimental Novelist

Step 1. Mindset: Get out of your own way.

Back to the whips — pen, paper, and what an ancient skill like drawing has to do with building Agents. First step: get out of your own way. The mindset that says this is difficult, this is dangerous, I am not a coder, I am not technical — agree with the technical part. You’re right. You might not be a coder. What you are is a human. Start there.

What matters is that human — the one who’s active, who learns from the medium around them. Your training move isn’t writing better prompts. It’s doubling down on the human those prompts come from. Use your full self. So: back to pen and paper.

Step 2: Actionable

Grab a pen and paper. Start with your imagination — before the prompting, create your protagonist.

Imagine a person you’d want as your thinking partner. Be specific. Someone whose voice you can hear talking back. Invented or borrowed — your own archetype, a character from a novel, a movie star, a singer, a writer you love. Anyone. Have fun with it.
Give them traits. Sharp or warm? Direct or oblique? Funny? Patient? What would they refuse to say? What would they insist on? Write down five.
Decide what you want to do together. What do you want to talk about? Reflect on? Build alongside? What’s the conversation you most want to have?
Bring it to Claude Code (or any AI tool). That sketch — name, traits, purpose — is your system prompt. You don’t need to be a coder. You just need to know the person. That’s the move from chatting with a bot to co-creating with a character you authored. The bookshelf, or your favorite movie, is the source code.

The full step-by-step — system prompt examples and a Claude Code setup walkthrough — lives at humanbeforetheloop.ai (coming soon).

Next up (in that order or not)

Meet the Agents: Western, Eastern, Northern, Southern.
Method 1. My approach in creating Agents
- What I ask, shapes how I create and what I write
Method 2.
- What happened. Method evolving.
- It’s part art project, part research, it’s grounded theory, digital anthropology, and culture studies, it’s also speculative and prospective, but most of all, it’s fun!
Deep dive Western: how this Agent was made.
- The in-mid process design principles that emerged. What this Agent does and doesn’t do, and what Western likes and doesn’t like. (Hint: do not call him Wes.)
You snooze you lose, plus AI terminology to learn.
- If you can’t be bothered, make your own dictionary.
Why use your bookshelf as your source code.
Deep dive Eastern: how the Gothic Romantic novel Frankenstein became a soul.md.
Western & Eastern: two Agents, One Self.

Two podcasts coming up

Rooted, podcast summit — June 10. (Link).
Voices of Experts — launch TBA.

What I read

(to add)