Editor’s pick
Start with this article
AgentsInternals
Coding agent harness: what I learned by cracking open Codex, OpenCode and Pi
Coding agents differ not by model but by the architectural bets of the harness. Reading the source of Codex, OpenCode, Pi and my own agent to tell load-bearing bets from ones you write to delete.
July 17, 2026
Worth reading
Strong articles that should not disappear in the archive
01 / 03
AgentsIndustry
Loop engineering: how to design loops for coding agents
August 17, 2026
InternalsPractice
02Why one character breaks the cache: prompt caching under the hood
Part two of the prompt caching series. We dig into the vLLM and paged attention source to find the byte-level reason beh...
June 26, 2026
AgentsBasis
03Prompt Caching: The Optimization That Fails Silently
Prompt caching cuts input token costs by 10x and keeps agent unit economics afloat, yet it breaks without a single error...
June 10, 2026
Topics
A quick way into the part of the archive you need
