You Don’t Need a Desk to Build

The desk was never the point

I’m writing this from my couch. Before that, it was my bed. Tomorrow it might be a coffee shop, or the airport, waiting on my flight to the next con.

Here’s the thing: I used AI more last month on my phone than I did from my desk. The dual monitors. The mechanical keyboard. The chair that cost more than my first car. Turns out, all of that was just ✨ aesthetic procrastination ✨ with better cable management.

The Lie We Tell Ourselves

“I’ll start that project when I have time to really sit down and focus.”

Sound familiar? Meanwhile, your phone sits in your pocket—a supercomputer capable of running AI assistants, connected to every repo you’ve ever touched, waiting for you to realize you were the bottleneck, not your environment.

The truth is brutal: the battlestation is a cope. A beautiful, RGB-lit cope that lets us feel like developers without actually developing.

My Setup (It’s Unserious and It Works)

This is what I use for my personal projects—the side quests, the experiments, the stuff I build because I want to, not because someone’s paying me:

Hardware:

• iPhone (yes, really)

• A computer under my desk (could just as easily be a cloud server)

Software:

• Happy Engineering app - Claude Code on my phone (requires a Claude Pro or Max subscription)

• PAI (Personal AI Infrastructure) - Daniel Miessler’s AI framework, tailored to my codebase, my preferences, and my neurodivergent brain. HUGE KUDOS for this. Absolute chef’s kiss.

• Git - Still git. Some things don’t change.

The computer’s under my desk (or in the cloud). I’m usually not.

PAI is the whole point.

I have my entire Personal AI Infrastructure around Claude Code. Skills that know how to do specific jobs. Memory that persists across sessions. Response formats designed for my neurodivergent brain. It remembers my projects, my preferences, my half-finished ideas from three weeks ago.

When I say “fix the research bug,” PAI knows which repo (usually HEARTH), which research system, and which bug I’ve been ranting about. I don’t re-explain context. I don’t copy-paste file paths. I just talk to it like a teammate who’s been in every meeting.

This isn’t autocomplete. It’s continuity.

The Part Where I Stopped Fighting My Brain

“Just sit down and focus.” Cool advice. About as useful as “just be taller.”

For years I thought productive meant more structure, more routine, more discipline. More of everything I was apparently bad at. I tried to force my brain into a shape it wasn’t designed for. Spoiler: it didn’t work.

Then I tried something different: meeting my brain where it actually is.

My brain wants to work at weird times. Fine. My brain can’t sustain attention for 4-hour deep work blocks. Fine. My brain gets distracted and needs to context-switch. Fine.

Mobile development doesn’t fight any of this. It leans in. Got 10 minutes waiting for food? Ship a fix. Lying awake at 2 AM with a solution? Actually implement it instead of praying you’ll remember it tomorrow. (You won’t. You never do.)

The Hemingway Technique

Ernest Hemingway famously stopped writing mid-sentence each day. Not because he was done, but because he knew exactly where to pick up tomorrow. The friction of “where do I even start?” vanished.

Here’s how it works for code:

Before Bed (5 minutes, from your phone):

1. Describe what you want to build

2. Let AI help you plan it

3. Identify the overnight task

4. Hit execute and go to sleep

While You Sleep:

• AI writes the code

• Runs the tests

• Documents what it did

Morning (whenever you wake up):

• Review the diff

• Merge or iterate

• Feel like a wizard

This isn’t theoretical. This post? Planned from my bed at 11 PM. Drafted by AI overnight. Reviewed and published while I had coffee.

“But what if the AI writes bad code?”

It might. You review it. Same as reviewing any junior dev’s PR, except this one doesn’t get offended when you request changes and actually learns from the feedback.

The Response Format That Changed Everything

I used to dread reading AI responses. Walls of text. Nested bullet points. Tables that didn’t fit on mobile. By the time I scrolled through the explanation, I’d forgotten what I asked.

So I built a format for my brain:

🎯 Task summary • Status

📌 RESULT:
Answer first. Always.

✅ DONE:
• What happened
• Evidence it worked

⏳ NEXT:
• What's coming

💬 PAI: 16 words max summary

15-20 lines max. Answer at the top. Emoji anchors for scanning. No tables.

The format got me to actually read the thing.

What I’ve Actually Built From My Phone

Because talk is cheap:

• Full PAI skill system - The AI infrastructure that powers my workflow

• Multiple HEARTH submissions - Threat hunting doesn’t care where you sit

• This entire blog setup - Hemingway’d into existence

• Bug fixes at 10 PM - The best time to fix bugs is when you’re annoyed enough to actually do it

• Automation scripts - Usually while waiting for something else

None of this required a desk. All of it required admitting that the desk was never the point.

The Obligatory “Be Smart About This” Section

Look, I’m not telling you to YOLO production deploys from your phone at 2 AM. (Okay, maybe I did that once. Don’t be like me.)

AI-assisted coding is powerful, but it’s not magic. You still need to:

• Review what gets generated - AI makes confident mistakes

• Understand what you’re shipping - Don’t merge production code you can’t explain

• Use sandboxing - Let AI run wild in containers, not on your host

• Know when to stop - Sometimes “wait until morning” is the right call

• Keep humans in the loop - Especially for anything that matters

The goal isn’t reckless speed. It’s removing artificial barriers between you and the work you actually want to do. Be smart. Be careful. But don’t let “being careful” become another excuse to never start.

Will this workflow change? Probably. The tools will get better, my needs will shift, and I’ll find new ways to break things. The principle stays: work where your brain works, not where productivity culture says you should.

Permission Granted

You don’t need the perfect setup. You need a phone and five minutes of “what if this works.”

You don’t need to fix your neurodivergent brain. You need tools that work with how you actually function.

Open your phone. Talk to an AI. Ship something.

The best time to build was when you had the “perfect” setup you never actually used. The second best time is now, from wherever you are, with whatever you have.

Resources

Tools Mentioned (and a bonus):

• Happy Engineering - Claude Code on mobile

• Claude Code - The AI foundation

• PAI (Personal AI Infrastructure) - My version of this system. Persistent memory. Custom skills. neurodivergent-friendly everything. This is the thing.

• Hemingway Technique - The bedtime planning workflow

• Superpowers - Jesse Vincent’s structured workflow system for AI agents - gives Claude Code guardrails and process

Related Dispatch Posts:

• “2026: The Year Builders Show Up” - More on the permission-to-build philosophy

• “Why You Should Build” - Lauren Proehl’s take on creating vs. consuming