21 posts
Two course cohorts, hundreds of student sessions and an agent pipeline: how reviewing Claude Code logs became RetroPair, the plugin I just open sourced.
Going solo at the Hack-Nation 6th Global AI Hackathon: the VC Brain challenge from Maschmeyer Group, learning venture capital from scratch in a day, and 12 features shipped through an SDD pipeline on Claude and Codex instead of a dev team.
The core artifact of a personal AI is your memory and skills - files under git. Swap the model and the harness; your memory moves with you.
In six months, skills turned a single Claude chat into my personal app store. How to replace SaaS with your own skills, and why it works beyond developers.
Spent a day breaking down Claude Code's enterprise settings. Two files, three delivery mechanisms, bypass gaps. Published a public repo.
Why 76% of AI agent tasks fail on the first try - not because of the model, but because of the harness - and what I keep in my stack.
Claude Opus 4.7: a half-day hands-on breakdown. New features, API breaking changes, effort levels, benchmarks, migration gotchas.
How to design a pipeline of autonomous subagents for mass prompt generation. 4 hours, 9 decisions, 8 bugs caught by an agent
Two weeks of research, 20+ frameworks in 6 categories - from BMAD to QuintCode. Comparison table, maturity model, expert quotes from practitioners.
How I processed 40,335 messages from a Telegram group using a swarm of 378 Claude Code sub-agents and got 2,764 structured entries
Testing 1M context in Claude Code on a real project - monorepo, long sessions, PR reviews. Numbers, nuances, limitations.
Tested the same 5-stage skill in Claude Code and Gemini CLI - results differ due to architectural constraints, not bad intentions.
How a community of developers and architects moved from AI skepticism to agentic flows in six months - documented with quotes
Agent Teams vs subagents - with prompts and real cases. When a team of 5 agents wins, and when it doesn't
How to build a Claude Code skill that converts PNG from Imagen into web-ready WebP with a single command
How real feedback from Telegram exposed CSS bugs I couldn't see myself - and why a fresh pair of eyes beats any QA process
A Claude Code agent sent an internal message to a public channel with 6k subscribers. Breaking down the incident and sharing a safety checklist for autonomous AI agents.
How I migrated prodfeat.ai from React to Astro 5 and boosted my SEO score by 34% using 10 AI agents
How to build a slash command for analyzing YouTube videos in Claude Code with n8n and Supadata API. Step-by-step breakdown with prompts and configs
How to build a working web app for video generation - with voice and prompts via Claude Code, no developers needed
Tested three frameworks for autonomous Claude Code - and realized the real problem isn't which tool you pick