Issue archive
Preview — not a sent issue
What an issue looks like
No issues have gone out yet. This is what this week's would carry.
Editor's note. Each issue opens with one failure mode from the week — what broke, why, and what to check in your own system.
This week's articles
- HarnessDev: A Model Can Write What Its Executor Reads
HarnessDev finds a self-built agent harness losing 36 points on SWE-bench Pro, or 20 on the authors' clean basis, when only the executor changes. What the executor reads transfers. What binds it does not.
- Where Jev Belongs in an Agent Harness (Not a Model Swap)
Jev's ordering is trustworthy; its confidence numbers are not. Where a System One decision model belongs in an agent harness, and where a threshold needs a local fit.
- Do You Need an Agent Framework? Start With Plain Code
State of the field as of 2026-09-18. Octomind ran LangChain's agent abstractions in production for more than twelve months, from early 2023 to mid-2024. When...
Get the next issue
One email a week, under five minutes to read.
Written by Ranjan Kumar — AI & ML Engineer · Author · Educator