Hi, I'm Bryce

I design AI systems that run real operations. Not demos. Production architectures with multi-agent coordination, quality gates, and self-improving memory. My flagship system manages an Etsy shop end-to-end with 34 specialized skills and three tiers of learned memory. I've evaluated and rejected two major orchestration frameworks against real requirements, and designed intent-based skill architectures that cut code 74% while increasing capability. Sixteen months and 260+ sessions of building, breaking, and rebuilding.

I find problems other people miss. I've diagnosed undocumented bugs in Google's Gemini SDK, Anthropic's desktop software ( startup freeze , file upload failure , 6 VM failure modes ), and Electron's multi-monitor handling. When standard troubleshooting fails, I trace root causes through log files, undocumented APIs, and system internals, then publish the fix so others don't hit the same wall.

I verify everything. Every blog post goes through a 4-agent review pipeline (fact-checker, technical reviewer, editorial reviewer, and visual QA) before it goes live. I've caught errors in my own published work, retracted incorrect findings, and corrected community misinformation. If I'm wrong, I say so.

Every project starts with a spec. I write detailed CLAUDE.md project files , some over 900 lines, that give every AI session the full context it needs without relying on conversation history. This means any session can pick up where the last one left off, even after a cold start.

I write intent, not instructions. My automation skills are markdown files that describe what should happen: goals, constraints, quality gates, and output contracts. The AI figures out the implementation. This approach cut my automation code 74% while making the system more capable, because clear intent scales better than detailed scripts.

Everything gets verified. Blog posts go through a 4-agent review pipeline before publishing. Business metrics get cross-checked against raw data sources. When my own analysis is wrong, and it has been, I catch and correct it before it becomes someone else's problem. I've retracted incorrect findings, corrected my own inflated metrics, and published corrections to community misinformation.

I kill what doesn't work. I've done a complete architectural rewrite in 6 days and abandoned 2 weeks of Chrome extension work after 8 PRs when a simpler approach was clearly better. Sunk cost isn't a reason to keep building.