Agent Infrastructure and Production RAG

Published on 07.05.2026

AI & AGENTS

Agentic Harness Engineering

TLDR: A raw LLM is just a next-token predictor. To turn it into a reliable agent you wrap it in a harness that controls execution, manages memory, and enforces safety boundaries.

Agentic Harness Engineering

Recursive Language Models

TLDR: When agents need to spawn sub-agents or break work down recursively, you need explicit design patterns to keep context coherent. Otherwise sub-tasks blow past the parent's intent and lose state.

Recursive Language Models

RALPH Loops

TLDR: RALPH loops are explicit execution patterns that force agents to reason reliably, finish tasks, and stop when they should. They are the antidote to wandering, never-terminating agents.

RALPH Loops

How I Automated 91% of My Business

TLDR: A transparent walkthrough of the workflows, agents, and pipelines used to run most of a one-person business on autopilot. The interesting bits are the seams between human and machine, not the models themselves.

How I Automated 91% of My Business

The Ultimate LLM Knowledge Base

TLDR: A personal "second brain" built by stitching Obsidian, Readwise, and NotebookLM into a retrieval system that synthesizes information instead of merely storing it.

The Ultimate LLM Knowledge Base

Agentic GraphRAG

TLDR: Standard RAG hits walls around context rot, data fragmentation, and unified memory. GraphRAG addresses these by treating retrieval as a data modeling problem with an explicit ontology, not a pure algorithm problem.

Agentic GraphRAG

Opik Case Studies: Shipping RAG with Weave CLI

TLDR: A walkthrough with Michael Maximilien on building Weave CLI and shipping production RAG to enterprise customers. The headline finding: a small open-source embedding model ranked around #130 on MTEB beat OpenAI by 11% on real customer data, because of a continuous stitch-evaluate-iterate loop.

Opik Case Studies: Shipping RAG with Weave CLI