QFM109: Machine Intelligence Reading List - April 2026

Source: Photo by Growtika on Unsplash

Two thoughts ran through this month's reading. The first was about what AI is doing to us: Wharton's Steven Shaw and Gideon Nave put a name and numbers on it in Cognitive surrender — when the AI is wrong, we go wrong with it, and we feel just as confident either way. Ethan Mollick's Sign of the future: GPT-5.5 maps the jagged frontier moving outward; Peter Marreck's Low-Hanging LLM Fruit lands the practitioner answer: LLMs aren't bad programmers, they're unchecked ones. Two pieces on Taste in the age of LLMs and What science fiction can teach us round out the philosophy column.

The coding-agent ecosystem had its busiest month yet. The coding-agent Cambrian catalogues six new frameworks (wuphf, wirken, gaia, botctl, GitNexus, spell) and Sebastian Raschka's anatomy that explains why they all look alike. Anthropic's Superpowers plugin landed for Claude Code, addyosmani/agent-skills and the Agent Skill Index collect production-grade skills, and MeisnerDan/mission-control brings command-and-control for solo founders running fleets of agents. Smaller pieces of the toolkit too: drona23/claude-token-efficient for terser responses, the .claude/commands/handoff.md gist, and disler/indydevtools and garrytan/gstack for opinionated setups.

Models and infrastructure: Pel — a programming language for orchestrating agents, Running Gemma 4 locally via LM Studio's headless CLI, omlx for Apple Silicon inference, Mediator.ai, the llm-wiki, and three from Anthropic: Claude Design, Project Glasswing for defensive cybersecurity, and Emotion concepts in an LLM. And brendanhogan/loophole: an adversarial moral-legal code system that compresses centuries of jurisprudence into a few minutes.

As always, the Quantum Fax Machine Propellor Hat Key will guide your browsing. Enjoy!

QFM109: Machine Intelligence Reading List - April 2026

Links

GitHub - brendanhogan/loophole: Adversarial moral-legal code system · GitHub

GitHub - addyosmani/agent-skills: Production-grade engineering skills for AI coding agents. · GitHub

GitHub - drona23/claude-token-efficient: One CLAUDE.md file. Keeps Claude responses terse. Reduces output verbosity on heavy workflows. Drop-in, no code changes. · GitHub

Low-Hanging LLM Fruit for Better Agentic Coding from Frontier Models

GitHub - MeisnerDan/mission-control: Open-source task management for the agentic era. The command center for solo entrepreneurs who delegate work to AI agents. · GitHub

Taste in the age of LLMs

.claude/commands/handoff.md

Running Google Gemma 4 Locally With LM Studio’s New Headless CLI & Claude Code

GitHub - jundot/omlx: LLM inference server with continuous batching & SSD caching for Apple Silicon — managed from the macOS menu bar · GitHub

Emotion concepts and their function in a large language model Anthropic

llm-wiki · GitHub

[2505.13453] Pel, A Programming Language for Orchestrating AI Agents

Introducing Claude Design by Anthropic Labs Anthropic

Cognitive surrender: Wharton names the thing we're all doing with our AI

The coding-agent Cambrian: six new frameworks, and the paper that explains why they all look alike

Mediator.ai

What Science Fiction Can Teach Us About Taste

Sign of the future: GPT-5.5

GitHub - garrytan/gstack: Garry Tan's Claude Code setup with 23 opinionated tools · GitHub

Project Glasswing: Securing critical software for the AI era Anthropic

Agent Skill Index

GitHub - disler/indydevtools: An opinionated, Agentic Engineering toolbox powered by LLM Agents · GitHub

Superpowers for Claude Code: a quick tour