// Writing Essays, guides & field notes
From the blog
Google Performance Max: How to Actually Control the AI Black Box (Asset Groups, Audience Signals, Budget Caps)
Performance Max campaigns burn budget when you can't see the levers. Here's how to actually control PMax using asset group structure, audience signals, exclusions, and budget caps — with real before/after numbers.
Build vs Buy vs Automate: A Decision Framework for 2026
The build-vs-buy binary is obsolete. This 5-factor scoring framework helps you choose between custom code, SaaS, and self-hosted automation — with real examples and a downloadable decision matrix.
Meta Ads MCP: What It Actually Lets You Do, How to Set It Up Safely, and Why I'm Not Trusting It With Live Spend Yet
A practical walkthrough of Meta's official MCP server for ads — what it actually lets you do with Claude and ChatGPT, how to set it up safely, and where conversational ad operations still break down.
n8n vs Zapier vs Make (2026): Pricing, Power, Self-Hosting, and AI Features — With a Clear Winner by Use Case
An opinionated, hands-on comparison of n8n, Zapier, and Make in 2026 — covering real pricing at scale, self-hosting, AI capabilities, and clear winners by use case.
I Open-Sourced Two Projects. Here's What Actually Happened.
I open-sourced two projects and tracked everything — stars, PRs, docs effort, and the silence after launch. Here's what actually happened and when open source is worth it.
AI UX Patterns That Actually Work in Production: Loading States, Streaming, Confidence, and Human Handoffs
Most AI products slap a chat UI on an API call and call it done. Here are the production-ready UX patterns that actually work — from streaming responses and confidence indicators to human-in-the-loop handoffs that preserve context.
Prompt Engineering Production Patterns That Actually Ship
Production LLM pipelines need more than zero-shot prompts. This guide covers four battle-tested patterns — few-shot, chain-of-thought, JSON mode, and tool-use — and shows how to combine them for reliable systems at scale.
Local-First Software Is the Future — Here's How I Built One With Tauri, SQLite, and CRDTs
Cloud-first apps break without internet and compromise your data. Here's why local-first architecture with Tauri, SQLite, and CRDTs is the better path — drawn from real experience shipping OpsConsole.
RAG Explained: Building AI Systems That Actually Know Your Data
Most RAG tutorials stop at the quickstart. Here's the opinionated guide to chunking, embeddings, vector stores, and retrieval that actually works in production — not just in notebooks.
SEO Isn't Dead: How to Rank in the Age of AI Search in 2026
SEO isn't dead — it's just playing by new rules. A practical playbook for ranking in Google AI Overviews, ChatGPT Search, Perplexity, and the zero-click reality of 2026.