Tech & AI Blog
This is where I write long-form about agentic engineering and generative AI. The posts here go deep: what is actually changing in how we build software, where the frontier labs are placing their bets, and what works (and what does not) when you put these tools into production.
If long-form is more than you have time for, I keep two shorter habits.
The Agentic Engineer Weekly is my newsletter, published every week and archived in its own newsletter section. One week, one read: what shifted and why it matters for builders.
I also post on LinkedIn almost every day. Shorter takes, faster reactions, more of the lived-experience side of the work. Follow me there if you want my thinking close to real time, between the long-form pieces here.
Everything below is the long-form archive only. Agentic engineering, MCP, model strategy, testing automation, and a few opinionated takes on how AI is reshaping the software engineering craft. Newest first.
For older content, mostly testing automation topics (Selenium, Playwright, BDD, QA frameworks) from 2018 to 2023, I used to write on Medium. That archive is still public if you are looking for legacy posts. I do not publish there anymore.
Evals Are Just QA With a New Name: How to Actually Test Skills and RAG
Every team hiring eval engineers already has the skill: it is called QA. This is the deep-dive companion to Episode 11, with a public repo you can clone: real code, real golden questions, and real numbers for evaluating the two things engineers actually build, Agent Skills and knowledge bases (RAG).
Read post
Your Second Brain Finally Gets a Standard: Inside Google's Open Knowledge Format
I run five LLM-maintained knowledge bases on a Mac mini, all built before the pattern had a name. In June, Google gave it a name and a spec: the Open Knowledge Format. This is a deep dive into OKF, the Karpathy LLM wiki it standardizes, how it relates to RAG, qmd, Obsidian, and graph layers, and the field-by-field audit that showed my hand-rolled knowledge base was one frontmatter field away from conformance.
Is My $200/Month AI Subscription Actually Worth It?
A few friends think I am wasting $200 a month on a Claude Code subscription. So I started keeping a ledger of what it produces. Here is the accounting model I am running, the worked maths, and the one return that never shows up on the invoice. Part 1 of a series.
AI Is Downstream of Context
Why connected tools beat smarter models in AI testing. Same model, same prompt, one 5 KB skill file of difference: one agent wrote a test that would re-introduce a fixed production incident, the other caught the ambiguity before writing a single test. The full A/B evidence, the reverse-engineering loop for legacy products, and the build/store/surface playbook.
Google I/O 2026 for Agentic Engineers: Seven Verticals, One Catch-Up Strategy
A postmortem of Google I/O 2026 filtered through the agentic-engineering lens. Every announcement that matters for builders, with my analysis on each. The thesis: Google is catching up on every vertical, and surface area is doing the rest.
Three Paths to Agentic UI Automation (and the One I'd Bet On)
Three ways an AI agent can drive a browser today. AI-generated code, live driving via MCP or CLI, or composed atomic tools. The math, the tradeoffs, and why letting the LLM click is the most expensive mistake in agentic UI automation.
Every Web Page Is Becoming a Function Call
Companion essay to Episode 03 of The Agentic Engineer. Web MCP, A2A, UCP, and a working bank you can clone tonight. The new browser primitive that quietly changes what UI automation means.
Ditch the Photographer, Embrace the AI: Generate Stunning Headshots for Free
Why pay for a professional photographer when AI can generate beautiful, varied headshots for the price of a coffee? A walkthrough of the workflow I used to ship my own.