Writing & Insights

Explorations at the intersection of AI practice and AI understanding

Patterns Over Systems: What Three Agent Projects Taught Me About Multi-Agent Work

April 18, 2026

The case that patterns matter more than infrastructure when you build with agents. Walks through three real projects, an optimization loop on the Rust TTS port, a generator and evaluator harness for iOS work, and the full governance system in Agent Vision Team, and pulls out the disciplines they share: files rather than chat as the communication channel, a fresh context on every spawn, hard quantitative gates, and a strict separation between doing and judging.

The Voice Mode Gap: One of AI's Biggest Opportunities

April 7, 2026

Voice modes on the major assistants still cannot create artifacts, produce documents, reach MCP servers, or do general tool use, even though the APIs underneath them already can. Whichever lab closes that gap first will have built something genuinely transformative, and it matters most for the people who rely on voice for accessibility.

Understanding-Based AGI: A Framework for Genuine Machine Intelligence

March 16, 2026

The primary introductory article for the knowledge distillation framework. Covers distillation versus compression, the dimensional world model, structural integration, grounding, and why genuine understanding requires fundamentally different architecture from current AI systems.

My Approach to Deploying OpenClaw: A Security-First, Multi-Instance Architecture on Proxmox

March 11, 2026

A security-first approach to deploying governed multi-agent coding architecture, covering multi-instance isolation, infrastructure decisions, and the practical realities of running autonomous coding agents responsibly.

Agent Vision Team: What Changed in Twelve Days

February 21, 2026

A follow-up documenting the rapid evolution of the Agent Vision Team system, what shifted in thinking and implementation over twelve days of intensive development.

Agent Vision Team: One Developer's Case for Governed Agentic Coding

February 9, 2026

A multi-agent system built on Claude Code to manage complexity in agentic development, enforcing vision, architecture, and quality standards through knowledge graphs and governance layers.

Using AI to Port an Incredible New TTS Model to iOS: PocketTTS for iOS

January 26, 2026

Documenting the process of porting Kyutai's Pocket TTS model to iOS using Rust and Candle, solving fourteen major technical issues with five specialized AI agents to enable on-device speech synthesis.

Crazy Speed w/o Limiting Features with AI-Assisted Development

January 8, 2026

An honest look at building enterprise-grade quality with a small team through Experience-Guided AI Development, layering multiple AI tools guided by deep human experience to approximate expert-led software development quality.

How I Built a Custom AI Curriculum in Minutes, and Why It Changes Everything About Learning

January 2, 2026

How AI enables custom and tailored, just-in-time knowledge distillation for me, and likely others.

A Random Evening's Exploration with AI: From a Simple Seed

September 17, 2025

A two-way conversation between human and AI, exploring how a minimal prompt can lead to thoughtful, surprising exchanges. An experiment in AI engagement that emerged from six months of intensive exploration.

AI Safety and the Need for Something More Fundamental

July 9, 2025

An under-informed practitioner's perspective on why AI safety needs fundamental, core-level constraints beyond surface training. Originally published on LinkedIn.

Note: I'm currently splitting my time between open-source AI projects and independent exploration of model architecture and AI safety. New writing is published on Substack. For updates, subscribe there or connect with me on LinkedIn.