Reading list
78 unread · 16 fetched and readable offline · built 6/4/2026
Up next
- Effective harnesses for long-running agentsanthropic.com
BUILD: Agent Harnesses
BUILD: Agent Harnesses
BUILD: Autoresearch
clean spectrum definition; reframes the landscape
EXPLORE: UX/interaction patterns for agent systems
PostHog: 5 hard-won lessons
Hamel Husain: harness IS data science
battle-tested practitioner advice
Sarah Wooders (MemGPT/Letta)
Harrison Chase: 3 layers of learning
EXPLORE: reframes skill systems (relevant to ~/.claude/skills)
failure mode taxonomy
Qwen team lead on agentic RL
LangChain on behavioral evals (feeds agent-series M07)
EXPLORE: open-source self-improving loop (feeds M08)
LangChain's hosted alternative
action-space construction (Anthropic CC insider)
best opener (per the learning map)
evals-driven harness improvement
comprehensive; the M1 cluster's anchor read
- Effective harnesses for long-running agentsanthropic.com
Anthropic eng
Anthropic eng
EXPLORE: novel layer
context efficiency (feeds M06)
Harrison Chase
pre-built configurable harness (on-thesis)
competitive landscape
Primary: 20 improvements, 11% speedup
Primary: 3-file architecture
best analytical breakdown
- Karpathy's Autoresearch for PMs: Complete Guidenews.aakashg.com
EXPLORE: non-ML applications
EXPLORE: extends to subjective domains, co-written with Hermes Agent
open-source Deep Research stack
- How we built our multi-agent research systemanthropic.com
Anthropic eng (feeds M09)
EXPLORE: swarm via tmux/worktrees
directly relevant to your setup
EXPLORE: LLM Council deliberation
Shpigford: USER.md + MEMORY.md (for David)
adversarial debate for sycophancy
SOUL.md/MEMORY.md/DREAMS.md (vault-as-product)
first-principles walk through agent memory
EXPLORE: agents forget like brains
where Claude-Mem sits in the agent stack
- GitHub - 1st1/lat.md: Agent Lattice: a knowledge graph for your codebase, written in markdown.github.com
EXPLORE: markdown knowledge graph + MCP server (directly vault-relevant)
server design, OAuth, vault patterns
Anthropic eng (feeds M04)
non-dev built full CoS in 36 hours (vault-as-product)
EXPLORE: 3-layer design harness for non-designers
- Prompt Engineeringlilianweng.github.io
Lilian Weng
- An update on recent Claude Code quality reportsanthropic.com
Anthropic incident postmortem (ops depth)
Unsloth practical guide
explainer for the MIT RLM paper
EXPLORE: Meta paper trains NN to simulate a computer
30 lessons (Holding + Game C)
EXPLORE
Dorsey + Botha: Block's AI-native org
a16z on building firms vs funds
Hashimoto on building blocks vs apps
EXPLORE: Hebbia CEO on org design
- Introduction - SITUATIONAL AWARENESS: The Decade Aheadsituational-awareness.ai
Aschenbrenner: AGI timeline (must-read)
Sequoia opportunity map (vault-as-product / services-as-software)
5 moats that survive AI
a16z
consumer-agent growth (feeds the consumer agent market report)
EXPLORE: services → product motion
- How to Do Great Workpaulgraham.com
Paul Graham
SF social capital, practical
SF landing (now: settling-in, not pre-move)
EXPLORE: Innovation/Capital/Deflation first principles
whole company on one laptop (vault-as-product)
Dan Koe on generalism (the compounder shape)
bootstrapping manifesto (Obsidian-as-company model)
EXPLORE: Yacine
"you can just build things"
Monte Carlo, Kelly, copulas with code
IR = IC * sqrt(N)
daily pre-market for Holding
- A decoder-only foundation model for time-series forecastingresearch.google
Google's TimesFM
Wiecki, intuitive Metropolis-Hastings
Wiecki, dependence beyond correlation