Skip to content

Instantly share code, notes, and snippets.

@Na0s
Na0s / Scaffolding Beats Sub-Agents: A Controlled Build Comparison.md
Last active April 23, 2026 13:18
Scaffolding Beats Sub-Agents: A Controlled Build Comparison

Three autonomous code-generation approaches implemented the same ~600-line Python project across three runs each: a solo Opus 4.6 agent, Opus 4.6 orchestrating Sonnet 4.6 sub-agents, and Opus 4.6 operating with a custom Agentic Project System (APS) skill set. All agents were provided the same fully specified plan document to work from. Solo Opus was fastest, cheapest, and had the lowest variance. APS produced the highest code quality at 51% higher cost. And Team delegation finished last on every measured dimension. At this task size, execution-structure scaffolding improves output even when working from a plan document.

Introduction

Autonomous AI code generation architectures are proliferating (single agents, team orchestration frameworks, full project pipelines) but empirical comparisons are thin on the ground. This study runs the same well-specified project through three build approaches and measures speed, cost, reproducibility, and code quality across three runs of each.

The three approaches tested

@Na0s
Na0s / Table of Abstracts: Applying Library Science to Create Lean Agentic Knowledge-bases.md
Last active April 21, 2026 22:47
Table of Abstracts: Applying Library Science to Create Lean Agentic Knowledge bases
title Table of Abstracts
tags
toa
tiered-retrieval
context-engineering
abstract Proposes a lightweight RAG alternative to vector databases. Per-file abstracts and per-folder overviews stored as YAML frontmatter, parsed by a script that extracts only the metadata. The AI retrieves context in tiers, reading only what it determines is relevant at each stage.

Editorial Note: