Tech Exploration & Product Thinking
Claude Code Writes 90% of My Code — But That's Not Its Most Important Job
Everyone talks about AI writing code faster. After 30 days running a one-person company with Claude Code, I found that coding is the least valuable thing it does.
I Shipped Two Products in 10 Days with AI. The Real Ceiling Was Not Building.
Two AI-powered MVPs, zero paying users. The bottleneck was not product quality — it was distribution. Real revenue data from solo founders reveals what actually determines your ceiling.
I Run My Entire Company From a Single Git Repo
No Zapier, no Notion, no n8n. One monorepo, one AI (Claude Code), and a 245-line CLAUDE.md file. Here's how I built an operating system for a one-person company: 10 agents, 10 commands, 139 commits in 5 weeks.
Google Traffic Dropped 33%, But Top Sites Grew — The Distribution Playbook for Zero-Audience Founders
Original research from 16 data sources: SEO is structurally declining, AI search cites only 2-7 sources per query. Here is a three-track distribution framework for founders starting from zero.
15,000 AI Products Are Fighting for Your Users — Here's Why 90% Will Die and What to Do Instead
The SaaSpocalypse wiped $285B from software stocks. Building costs collapsed from $100M to $30. After two failed products, I found the survival playbook for indie developers.
An 18K-Star Agent Orchestrator, and Why You Probably Don't Need It
ruflo coordinates 60+ specialized Agents with Byzantine fault-tolerant consensus and Q-Learning routing — serious enterprise-grade orchestration. But for most AI developers, a single Agent is the right starting point.
The 20-Person Company Is Disappearing — Which Company Form Fits the AI Era?
AI isn't making every company smaller — it's hollowing out the middle. From solo AI fleets to micro elite teams, which configuration fits you? Includes a product form x company form selection matrix.
Two Products in 10 Days, Zero Users — My First Loop as a Solo Founder
I used an AI-powered demand mining system to pick ideas, built two MVPs in 10 days, and both failed validation. A full postmortem of BuyOnce and Traction, plus the 3 rules I hardcoded into my system afterward.
I Built a SaaS That AI Could Clone in an Afternoon — Then I Figured Out Why
I built a SaaS and realized AI could replicate it in an afternoon. The problem wasn't the features — I was playing on the wrong layer. This forced me to map out a "Stack Ownership Spectrum," and led to a counterintuitive discovery: going lighter isn't safe either.
5 Architecture Decisions Worth Stealing from an 18K-Star AI Agent Toolkit
A deep dive into 5 architecture decisions from pi-mono: runtime provider registration, dual-layer messaging, injectable tool operations, diff rendering, and session trees. Each one is ready to use in your own AI agent project.
Your AI Conversations Break After 20 Turns? You Need These Three Lines of Defense
Distilled from OpenClaw's context management system: a three-tier defense strategy against context window overflow in LLM applications
Stop Hardcoding Your System Prompts — A Prompt Assembly Pattern Extracted from 672 Lines of Code
A modular, conditionally-rendered prompt assembly framework distilled from OpenClaw's system prompt builder
I Read 1 Million Lines of Code and Found the Layer Most LLM Apps Are Missing
Distilling production-grade resilience patterns from OpenClaw open-source Agent engine into a reusable library for LLM applications
Deep Dive: Comparing the Big Four LLM APIs — ChatGPT, Claude, Gemini & Grok
A systematic comparison of session-layer design across ChatGPT, Claude, Gemini, and Grok — message models, context management, standout features, API styles, and the divergent product philosophies behind them