Weekly Log
Week 38
Sep 14–20, 2026
Local LLM experiments
Set up Ollama and connected local models to VS Code through Continue. Tested coding workflows and hit the practical limits of long context and CPU-only inference.
Started expanding Far Away from Codex
Moved the project beyond simple completion notifications toward an attention layer for coding agents: approvals, input requests, failures, and bounded remote responses.
RepoRead through ACP
Studied Agent Communication Protocol and compared where a standard protocol makes sense versus native agent hooks.
SourceKaggriculture experiment
Tried an agent/model approach for the Kaggriculture competition. The initial results weren’t promising enough, so I stopped working on it.