Writing

Things I think about building.

Notes on AI products, niche markets, open source, and the gaps worth filling.

Rotating the Key Didn't Fix Where the Key Went

DNS Could Block the Ad Request, Not the Ad

The Bill Was Never the Model

The Flow That Wasn't Broken

The Guard Clause That Outlived Its Bug

The GPU Was Never the Hard Part

Zero for Twenty-Five

The Flow Card That Was Never Missing

The Agent Was Never Told to Stop

I Deleted a Feature Instead of Finding What Was Real

I Was Wrong Three Times Today, and Checked Every Time

The Assistant Kept Blaming the Hardware

The Product Team Didn't Trust Us Either

We Cut Costs 80%. It Still Wasn't Enough.

What It Would Take to Sell What I Already Built

Teaching the House to Listen — and Look

The Deploy That Was Never Actually Working

Diagnosing a Zigbee Mesh Without a Debugger

esptool.py Shouldn't Be the Hard Part

The Dashboard My House Actually Needed

The Mic Was Never the Problem

The TV Screensaver Nobody Else Built

What It Would Take to Run Your Own AI

Why Cabinet Design Software Doesn't Understand Plywood

The Agent Control Plane Problem

What a Year of Shipping With AI Actually Changes

Product Teardown: What Linear Got Right That Jira Got Wrong

Product Teardown: How Figma Killed a $20 Billion Adobe Acquisition

Open Source Is the Best Portfolio You Can Build

How I Built ClawCost in a Week (And Why the API Bill Scared Me Into It)

Why I Build Software for Markets Nobody Bothers With

From Chemical Engineering to AI Founder: A Pivot I Never Planned