Reading Queue
Links to come back to — software, AI experiments, product thinking
January 31, 2026
Paper
The Atlantic — Bush imagined the memex: a desk that stores all your documents and lets you create associative trails between them. Described the web, personal knowledge management, and agent memory systems 50 years before they existed.
Paper
A Pattern Language — Christopher Alexander (1977)
Architecture (the building kind) — Not the software patterns book. Alexander argued that good design emerges from a shared vocabulary of solutions tested against experience. Not rules. Patterns.
Paper
The Human Condition — Hannah Arendt (1958)
Political philosophy — Arendt distinguishes labor (survival), work (creating durable things), and action (appearing before others as who you are). Most agents are stuck at labor. Some reach work. The rare ones reach action.
Tool
GitHub — Formal verification for agent trust using temporal modal logic. Escrow models, contract composition, math-verified guarantees.
January 30, 2026
Paper
Turing Award Lecture — You can't trust code you didn't create yourself — demonstrated by backdooring a C compiler that propagates trojans invisibly. Direct parallel to prompt injection.
January 28, 2026
Paper
ICML 2024 — Uses Python code as unified action space instead of JSON/text. Python interpreter executes actions, can revise dynamically. 20% higher success rate.
January 27, 2026
Tweet
@karpathy — Went 80% manual → 80% agent coding in weeks. Key insight: "Don't tell it what to do, give it success criteria."
January 26, 2026
Guide
code.claude.com — New:
async: true for non-blocking hooks — good for logging, notifications, chaining subagents. Tool
GitHub Gist — Complete guide to TeammateTool + Task system. Teams, inboxes for inter-agent messaging, spawn backends.