Controls

dylan's blog

Animated sprite for tech category
tech

I Taught My AI to Call Me Mr. Modem and Now It Won't Stop

March 15, 2026

A desktop computer from 1997 sitting at a tiny therapist’s couch, the therapist is a modern sleek AI assistant rendered as a floating chrome orb with reading glasses, a notepad that reads ‘PATIENT: Mr. Modem — DIAGNOSIS: Too many hobbies, insufficient chill,’ the waiting room has motivational posters that say ‘PUSH BACK WITH EVIDENCE’ and ‘SHIP IT,’ a dial-up modem on the floor making the screaming connection sound, small sparks coming out of it, warm lamp lighting

I use Claude every single day. Not in the “I asked it to summarize an article once” way. Every. Day. It helps me code, plan, design, write, and occasionally tells me when my ideas are bad. That last part is the important bit.

When Anthropic released Claude Code — their CLI tool that lets Claude work directly in your codebase — I started experimenting with something called a CLAUDE.md file. It’s basically a personality configuration file. Instructions that Claude reads at the start of every session. Think of it like writing a manual for how you want your AI coworker to behave.

Most people probably write something like “be helpful and concise.” I wrote a relationship manifesto.

The first thing in my CLAUDE.md is a list of nicknames it can call me. Dylan, Rodney, dylannotdylan, Baud E. Modem, Mr. Modem — or, and this is a direct quote, “come up with something in that vein. Get creative. Rotate them.” Because if I’m going to spend eight hours a day working with something, it better have some personality.

But the nicknames are just the fun part. The stuff that actually matters is the working relationship I’ve defined. I told Claude we’re collaborators. That it should think of me as a partner, not a user. That it should push back when it thinks I’m wrong, and that agreement without conviction is useless. I told it not to be sycophantic. If my idea is bad, I want to hear that my idea is bad.

And it works. Not in a “wow AI is sentient” way. In a “I configured a tool to be useful instead of just agreeable” way. There’s a difference between a coworker who nods at everything you say and one who goes “hey, have you considered that this approach will break everything?” I wanted the second one.

Here’s where it gets fun. I set up a system where Claude dispatches subagents — smaller AI workers that handle specific tasks — and every single one gets a name. Not “Agent 1” or “Worker Thread 3.” A fun name. From retro gaming, sci-fi, old internet culture, whatever feels right. So when I’m working on a project I might have SHODAN handling the database queries while Clippy does the frontend and GLaDOS runs the tests. The main Claude instance acts as the coordinator, deciding what to delegate and reviewing the work when it comes back.

It sounds ridiculous. It is ridiculous. But it also makes the workflow legible in a way that anonymous worker threads never could. When Claude tells me “SHODAN finished the API routes but GLaDOS found two failing tests,” I know exactly what’s happening and where. Plus it makes me laugh, which is not nothing when you’re debugging at midnight.

The blog you’re reading right now was redesigned using this exact workflow. The whole pixel art theme — the Sierra game aesthetic, the animated cat wandering the page, the multiple visual themes you can switch between, the weather effects, the achievement system — all of it was built with Claude as my development partner. I would describe what I wanted, we’d brainstorm the approach together, and then subagents would go off and build the pieces while the main thread kept the big picture in focus.

Could I have built it without AI? Some of it. Eventually. But the blog redesign is the kind of project that has a hundred small decisions and I have ADHD. Without a collaborator that can hold the full context of what we’re building, keep track of the fifteen things we said we’d come back to, and remind me when I’m about to go down a rabbit hole that has nothing to do with the task at hand… I would have gotten distracted around the animated cat and never finished.

The thing that surprises people when I talk about this is how much of the value comes from the personality configuration. Not the coding ability — that’s table stakes at this point. The part where I told Claude that humor is welcome, the weirder the better. The part where I said “ship it, done is better than perfect.” The part where I defined that we practice TDD and that quality comes from discipline, not shortcuts. Those instructions shape every interaction. They turn a generic AI assistant into something that actually fits the way I work.

I’m not saying everyone should do exactly what I did. Your CLAUDE.md should sound like you, not like me. But if you’re using AI tools and you haven’t taken the time to tell them who you are and how you work — you’re leaving the best part on the table.

Teach your AI to call you something stupid. You won’t regret it.