Blog
Notes on building for the web and mobile: performance, architecture, AI, and the craft of shipping.
Agent loop
Building a coding agent from scratch, one loop at a time, every part in runnable code.
- 01What an agent loop is (and isn't): state, action, stop
- 02The simplest loop that works: a write-test-fix agent, step by step
- 03The sandbox in an agent loop: from the model's text to a reliable observation
- 04Who writes the agent's tests: three ways to define what's correct
- 05From generating files to using tools: a code agent's ReAct loop
- 06The agent inside a real repo: isolating tasks with git worktree and shipping a PR
- 07From a Jira ticket to a pull request: the end-to-end autonomous system
- 08The hard limits: what keeps it from disasterComing soon
Data annotation jobs: what labeling and reviewing AI responses involves
What data annotation jobs and AI response review are: the tasks involved, how they feed model training (RLHF), and how annotator quality is measured.
Orca: the ADE for orchestrating a fleet of coding agents in parallel
I tried Orca, the open-source Agent Development Environment (ADE), to run a fleet of coding agents in parallel—Claude Code, Codex and more, each in its own isolated worktree.
Model routing with OpenRouter and DeepSeek to cut costs without losing quality
How I routed models in production with OpenRouter, using DeepSeek for the bulk of traffic and premium models only where it matters, cutting LLM costs without degrading quality.
I automated a news site with n8n: from 20 RSS feeds to WordPress drafts
How I built an n8n pipeline that reads 20 RSS feeds, deduplicates with Postgres, curates and rewrites with tiered AI, and leaves the article as a WordPress draft with an image and a Telegram alert.
