Writing
Blog
Writing on production AI systems, LLM architecture, and AI agents.

I tested whether AI PR reviews catch cleanup I never asked for
I ran 48 reviews with a free code model and Claude Opus 5.5 to test unrequested cleanup. Both caught the reorder, but some findings invented requirements.

How I set up PR reviews for vibe coding with Kody and OpenRouter
My Kody and OpenRouter setup for reviewing AI-written code: BYOK pricing, model routing, real dashboard screenshots, and what free-model fallbacks can and cannot do.

Grok Bot vs OpenClaw
Hosted Grok versus self-hosted OpenClaw for beginners who want to automate work. Both sides, no pick — start free, upgrade when a wall shows up.

Grok Bot vs Claude Desktop
A both-sides compare of conversational Grok and Claude Desktop for beginners who want to automate work. No pick, no eval table.

Claude Desktop for Beginners
Install Claude Desktop, start a first chat, add a novice MCP connection, and keep secrets out of the prompt. Not Claude Code.

Grok 4.6 costs, limits, and evals vs the models I actually pay for
Grok Build is the coding-agent product. Grok 4.6 is the model that powers it. Official evals, API costs, and limits — copied from xAI, not rounded.

How I Actually Use Grok Bot
I use conversational Grok as a thinking, research-preparation, and drafting tool—not as an authority. Here is the six-step workflow I actually follow, including how I verify claims and protect sensitive data.

How AI Works: From Data to Decisions — A Developer's Introduction to AI Systems
A developer's guide to how AI works end to end: defining the decision, preparing data, training or selecting a model, running inference, applying deterministic product controls, and monitoring real outcomes in production.

Why Most AI Projects Die Before Production (And It's Not a Tech Problem)
The most-cited AI failure stat is made up, but the real numbers are worse, and the reason is almost never the technology.