Writing
Series to read in order, then single posts from real builds: what broke, what it cost, and the number that settled it.
Series
3 series
A few posts on one subject, written to be read in order.
Series · 8 parts
Testing AI agent security
How agent harnesses decide who each message speaks for, and what that does to prompt injection, measured on real APIs and a real email agent.
Series · 6 parts
How an LLM runs
What happens inside a language model when it answers you, one step at a time, measured on a real model.
Series · 2 parts
MCP in practice
How MCP works, what I learned building and using MCP servers, and how local and remote servers connect, step by step.
Posts
8 posts
What broke, what it cost, and the number that settled it.
Post · 21 Oct 2026
Re-checking 413 web pages with Jev: 23 needed a full read
How affiliateprogramterms.com keeps its terms current: one Jev yes/no request per page, $0.04 a refresh, and a test against Claude Haiku and Opus on 66 pages.
Post · 15 Oct 2026
Jev 1.14 vs OpenCV YOLOX: image questions and bounding boxes
I tested Jev 1.14's image input and new Box answers on 6 photos and compared the boxes with OpenCV's YOLOX detector. Accuracy, box fit, speed and cost.
Post · 3 Oct 2026
How I calibrated an LLM judge to grade like me, 25× cheaper
Evals on 188 answers from ChatGPT, Claude, Chatbase and my own pipeline over product datasheets, and a hybrid judge with Jev that grades like I do.
Post · 2 Oct 2026
Can an AI agent submit Upwork proposals? Yes, with your OK
Upwork's official MCP lets Claude, Codex or Cursor send a proposal for you. What it shows before sending, what stays on upwork.com, and what I got wrong.
Post · 20 Sep 2026
Ask twice: seven measurements from building with Jev
A coordinate lookup scored 3/24 where handing the model the naming job scored 22/24. Then two claims I had argued rather than measured turned out to be wrong.
Post · 19 Sep 2026
Choice, Score and Noul: four mistakes with Jev's primitives
I built a Magic 8 Ball on Jev, TypeSafe's calibrated-decision model, and misread its three primitives four times. The measured numbers, and what each mistake cost.
Post · 4 Sep 2026
AI turns emails into spreadsheet rows. Did it get them right?
Eight real inbound emails through Claude Haiku: 61 of 64 fields correct, every miss on one field. What a failure taxonomy shows that an accuracy score hides.
Post · 17 Jun 2026
98.3% vs 96.3%: the cost of an auditable AI extraction pipeline
105 labeled invoices and receipts: Claude Opus vs Haiku+Sonnet, a 2% accuracy gap and a 12× cost gap, plus the eval set and audit log behind the numbers.