Reading list

129 unread · 19 fetched and readable offline · built 5/9/2026

BUILD: Agent Fundamentals

Read these first if new to agents. Foundational patterns + 2026 landscape overviews.

Why We Thinklilianweng.github.io
Lilian Weng on inference-time reasoning, chain-of-thought, scaling test-time compute
Difference Between Harnesses & Frameworksx.comexternal
clean spectrum definition; reframes the landscape
Scaling Managed Agents: Decoupling the brain from the handsanthropic.com
Anthropic eng (skim — already discussed)
The 2026 AI engineer roadmapx.comexternal
"tutorial hell is comfortable; world demands systems" — opinionated 2026 stack overview
What to Learn, Build, and Skip in AI Agents (2026)x.comexternal
opinionated triage of the 2026 agent landscape
AI Agent Stack Everyone Must Use in 2026x.comexternal
builder's guide to agent stack components
Designing for Agentsx.comexternal
UX/interaction patterns for agent systems
The era of single-model engineering is overx.comexternal
multi-model routing as default, not afterthought
What we wish we knew about building AI agentsx.comexternal
PostHog: 5 hard-won lessons
The Revenge of the Data Scientistx.comexternal
Hamel Husain: harness IS data science
How To Be A World-Class Agentic Engineerx.comexternal
battle-tested practitioner advice
Why memory isn't a pluginx.comexternal
Sarah Wooders (MemGPT/Letta)
Continual learning for AI agentsx.comexternal
Harrison Chase: 3 layers of learning
Skill chaining and why skills should be actionsx.comexternal
reframes skill system
How To Solve Problems Of Long Running Workflowsx.comexternal
failure mode taxonomy
From "Reasoning" to "Agentic" Thinkingx.comexternal
Qwen team lead on agentic RL
How we build evals for Deep Agentsx.comexternal
LangChain team on behavioral evals
The Great Convergencex.comexternal
companies converging on self-improving agents
auto-harness: Self-improving systemsx.comexternal
open-source self-improving loop
How My Agents Self-Heal in Productionx.comexternal
self-healing deploy pipeline

BUILD: Agent Harnesses

W18+ priority. Want to build with OpenClaw and/or Hermes. ~20 articles, dominant theme of W17 X-triage.

What Hermes Agent Can Do for Youx.comexternal
blocmates intro to Hermes
Deep Agents Deployx.comexternal
LangChain's hosted alternative
Building Data Agents with Deep Agent SDKvorashm.substack.comexternal
practical SDK walkthrough
Lessons from Building Claude Codex.comexternal
action space construction
How I built harness using Claude Code leaksx.comexternal
direct reverse-engineering walk
A Closer Look at Harness Engineering from Top AI Companiesx.comexternal
synthesis of OpenAI/Cursor/Anthropic blogs
The Harness Is Everythingx.comexternal
best opener (per the learning map)
Harness Engineering Is Cyberneticsx.comexternal
control theory framing
Sandboxes Are the Servers of the Harness Erax.comexternal
backend metaphor
The Harness Is the Backendx.comexternal
same metaphor, different angle
Every Harness Will Become A Clawx.comexternal
calcsam vs dexhorthy debate
Agent Harnesses Are Dead. Long Live Agent Harnesses.x.comexternal
frameworks vs harnesses
harness as new community languagex.comexternal
three layer model (weights → context → harness)
Better Harness — Hill-Climbing with Evalsx.comexternal
evals-driven harness improvement
The Harness Stack — Layered Modelx.comexternal
synthesis of Q1 2026 OpenAI/Cursor/Anthropic reports
The Definitive Guide to Harness Engineeringx.comexternal
Systems Engineering Building Agentic Softwarex.comexternal
Bell Labs framing
Effective harnesses for long-running agentsanthropic.com
Anthropic eng on patterns for sustained agent work
Harness design for long-running application developmentanthropic.com
Anthropic eng on production harness architecture
The Self-Healing Agent Harnessx.comexternal
Catalyst — Monitor, train, deploy self-improving AI modelsx.comexternal
Context providers — missing layer between agents and toolsx.comexternal
The Hidden Cost of Fewer Tokensx.comexternal
context efficiency in Playwright CLI
Your harness, your memoryx.comexternal
Harrison Chase
Effective context engineering for AI agentsanthropic.com
Anthropic eng: definitive context engineering reference
Launching Claude Managed Agentsx.comexternal
pre-built configurable harness
Agents SDK long-running agentsx.comexternal
sandboxes, controllable harness, memory control

BUILD: Autoresearch

You saved this topic 8 times. Read the primaries.

Thread by @karpathyx.comexternal
Primary: 20 improvements, 11% speedup
GitHub - karpathy/autoresearch: AI agents running research on single-GPU nanochat training automaticallygithub.com
Primary: 3-file architecture
How Karpathy's Autoresearch Worksx.comexternal
Best analytical breakdown, 5 lessons
Karpathy's Autoresearch for PMs: Complete Guidenews.aakashg.com
Non-ML applications
How I used autoresearch to fix Gumroadx.comexternal
production proof
Applied Autoresearch to LLM Inferencex.comexternal
MLX optimization results
Autoreasonx.comexternal
extends AutoResearch to subjective domains, co-written with Hermes Agent
How to build a Deep Researcherx.comexternal
open-source Deep Research stack beating OpenAI/Gemini/Perplexity

BUILD: Multi-Agent & Orchestration

How we built our multi-agent research systemanthropic.com
Anthropic eng on the Research feature's multi-agent architecture
How we built LangChain's GTM Agentx.comexternal
production agent, 250% conversion lift
OpenClaw + Codex Agent Swarmx.comexternal
Swarm via tmux/worktrees
The Setup That Saved Me Hoursx.comexternal
multi-agent coordination
5 Claude Agents, 5 Discord Channelsx.comexternal
directly relevant to your setup
How to build an AI Agent company using Paperclipx.comexternal
CEO/worker hierarchy
Agreement is a bug (11 agents)x.comexternal
LLM Council deliberation

BUILD: Agent Memory & Personalization

How I built a hyper-personalization systemx.comexternal
Shpigford: USER.md + MEMORY.md (for David)
@karpathy AutoReasonx.comexternal
adversarial debate for sycophancy
A Theory of Mind in Three Filesx.comexternal
SOUL.md/MEMORY.md/DREAMS.md
Build Agents that never forgetx.comexternal
first-principles walk through agent memory
How AI Actually Remembersx.comexternal
full guide on why agents forget + how to fix
Geometry of Forgettingx.comexternal
memory systems for LLMs forget exactly like brains
Latent Briefing — KV Cache Compactionx.comexternal
multi-agent memory sharing
The Cognition Layerx.comexternal
bottom-up map of where Claude-Mem sits in the agent stack

BUILD: MCP & Tools (MCP-proper only)

See `research/reading-notes/build-agent-context-and-tools.md` for the filesystem-as-tools split that came out of W16-W17 reading.

GitHub - 1st1/lat.md: Agent Lattice: a knowledge graph for your codebase, written in markdown.github.com
markdown knowledge graph + MCP server
Building agents that reach production systems with MCPclaude.com
server design, OAuth with CIMD and vault patterns
Writing effective tools for AI agents—using AI agentsanthropic.com
Anthropic eng: tool design principles for agent reliability

BUILD: Practical Setups

My chief of staff, Claude Codex.comexternal
non-dev built full CoS in 36 hours
I Built an AI Company with OpenClawx.comexternal
closed-loop agent architecture
Design Without Designingx.comexternal
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 on a model degradation root cause

BUILD: AI/ML Deep

Papers + technical posts on RL, neural computers, model architecture. Slower reads, lower priority than agent harnesses for active focus.

Reinforcement Learning environments and how to build themunsloth.ai
Unsloth practical guide
Agentic RL — Decoupling Reasoning from Tool-Usex.comexternal
Recursive Language Models From Scratchx.comexternal
implementation walkthrough
Recursive Language Models by MIT, Clearly Explainedx.comexternal
explainer for the MIT RLM paper
Neural Computers, Explainedx.comexternal
Meta paper trains NN to simulate computer

FOUNDER/LIFE: GTM & Distribution

Granola $0 → $1.5B in 3 yearsx.comexternal
growth playbook
The 5-Step Outbound Systemx.comexternal
itsalexvacca, 300+ companies experience
5 levels of AI marketingx.comexternal
shannholmberg
5 pipelines I'd sell today using Claude Code (none are coding)x.comexternal
sellable workflows
How To Build Own Content Enginex.comexternal
DeRonin runs 10 socials, no manual posts
Thread by @yasser_elsaid_x.comexternal
Chatbase $9M ARR playbook
Thread by @gregisenbergx.comexternal
agency-to-SaaS playbook
Greg Isenbergx.comexternal
agency-to-SaaS $10M exit math

FOUNDER/LIFE: Org & Capital

Think Like a Family Officex.comexternal
30 lessons. Game C aligned.
YC on building AI company from ground upx.comexternal
From Hierarchy to Intelligencex.comexternal
Dorsey + Botha: Block's AI-native org
Firm > Fundx.comexternal
a16z on building firms vs funds
The law firm modelx.comexternal
pod model for AI agencies
The Building Block Economyx.comexternal
Hashimoto on building blocks vs apps
Thread by @NoahEpstein_x.comexternal
PE firms as ideal AI niche
Productive Individuals Don't Make Productive Firmsx.comexternal
Hebbia CEO

FOUNDER/LIFE: AI-Native Strategy

Why Your "AI-First" Strategy Is Probably Wrongx.comexternal
99% AI code, what they learned
Services: The New Softwarex.comexternal
Sequoia opportunity map
The AI Businesses That Will Print Money in 2027x.comexternal
what nobody is building yet
How to Build Services-as-Software Businessx.comexternal
itsalexvacca on the Sequoia thesis
The "Mismanaged Geniuses" Hypothesisx.comexternal
models good enough, the bottleneck is elsewhere
There are only two paths left for softwarex.comexternal
a16z
The Only Moats That Matterx.comexternal
5 moats that survive AI
Introduction - SITUATIONAL AWARENESS: The Decade Aheadsituational-awareness.ai
Aschenbrenner: AGI timeline
Thread by @rahulgsx.comexternal
Ramp VP: bitter lesson for agents

FOUNDER/LIFE: Identity & Life

How to Do Great Workpaulgraham.com
Paul Graham
The SF Group Chatx.comexternal
daniel_dhawan, SF social capital practical
So you moved to SFx.comexternal
practical for May 15 move
Number Go Downx.comexternal
Innovation/Capital/Deflation first principles
The solo founder stack of 2026x.comexternal
whole company on one laptop
If you have multiple interestsx.comexternal
Dan Koe on generalism
Thread by @levelsiox.comexternal
bootstrapping manifesto
being a human in the singularityx.comexternal
Yacine
Martin's Health Dashboardhealth.martinamps.comexternal
"you can just build things"

FINANCE: For Holding, David, Atlas

Quant models, financial AI, investment strategies

Comprehensive Guide to AI in Financial Researchx.comexternal
Daloopa CEO
How to Simulate Like a Quant Deskx.comexternal
Monte Carlo, Kelly, copulas with code
The Math Behind Combining 50 Weak Signalsx.comexternal
IR = IC * sqrt(N)
Game Theory on Polymarketx.comexternal
Kelly, Bayesian, Nash with code
How to Set Up an AI Stock Screenerx.comexternal
Daily pre-market for Holding
Thread by @mrjainx.comexternal
Claude wealth management plugin
The algorithm that crashed Wall Streetx.comexternal
TurboQuant economics
A decoder-only foundation model for time-series forecastingresearch.google
Google's TimesFM
All-AI hedge fundx.comexternal
2.55 Sharpe, 2.79 Sortino, $50M+ capacity, no human PMs
Don't listen to the peanut gallery; USVC is a watershed momentx.comexternal
private markets liquidity (Atlas-relevant)

STATISTICS: Foundational Methods

Probabilistic thinking, sampling, dependence structures. Feeds quant work + JL/TurboQuant lineage.

while my_mcmc: gently(samples) - MCMC sampling for dummiestwiecki.io
Wiecki, intuitive intro to Metropolis-Hastings
while my_mcmc: gently(samples) - An intuitive, visual guide to copulastwiecki.io
Wiecki, dependence beyond correlation