Learning Resources

Standalone interactive pages for exploring ideas and concepts.

2026-07-04

How to Train Your Agent: Building Reliable Agents with RL — Kyle Corbitt, OpenPipe

Interactive breakdown of Kyle Corbitt's (OpenPipe) AI Engineer World's Fair talk: how a 14B open model trained with reinforcement learning (ART·E) beat o3 at email research — and the four-step recipe to do it yourself. Start with a prompted baseline, build a realistic environment, design a reward function, and survive reward hacking.

Open →

2026-07-02

How I Deleted 95% of My Agent Skills and Got Better Results — Nick Nisi, WorkOS

Interactive breakdown of Nick Nisi's (WorkOS) AI Engineer Summit talk: why your agent is lying to you, why deleting 95% of your skills made things better, the gotcha-only 553-line alternative to a 10,000-line doc-dump, and why every failure is a harness bug — not an agent bug.

Open →

2026-06-30

Dynamic Subagents — Running Parallel Agents Reliably (LangChain Deep Agents)

Interactive breakdown of LangChain's Dynamic Subagents: why orchestrating subagents in the agent's head breaks at scale, how writing code (the eval tool + task global) fixes it, and the six workflow patterns — Classify & Act, Fan Out & Synthesize, Adversarial Verification, Generate & Filter, Tournament, and Loop Until Done.

Open →

2026-06-30

How We Build Effective Agents — Barry Zhang, Anthropic

Interactive breakdown of Barry Zhang's (Anthropic) AI Engineer Summit talk: the spectrum from features to workflows to agents, when NOT to build an agent, keeping agents simple (models using tools in a loop), thinking like your agent, and the open questions ahead.

Open →

2026-06-24

Attribution Modeling, Visualized — Who Gets Credit for the Sale?

Interactive explainer of marketing attribution models: switch between first-touch, last-touch, linear, time-decay, position-based (U/W/Z), and data-driven and watch how each splits the credit for one conversion across a customer journey — plus how to choose a model and why none is ever 100% accurate.

Open →

2026-06-23

How the Hermes Agent Works — Components & Architecture

Interactive guide to Nous Research's open-source Hermes agent: one agent + one memory across every surface, its six capabilities (Connect, Remember, Schedule, Delegate, Search, Experiment), persistent skill-learning memory, isolated zero-context-cost subagents, gateway scheduling, and the five execution backends.

Open →

2026-06-17

How Auth Brokers Work — Connecting Services with Stytch & Composio

Interactive guide to auth middleware: how a service like Composio or Stytch brokers OAuth between apps — the four roles, the authorization-code flow, connected accounts, automatic token refresh, and consumer vs. provider architectures.

Open →

2026-06-12

TurboQuant: Vector Quantization for LLM Efficiency

Learn how TurboQuant compresses high-dimensional vectors using two-stage quantization and random rotation, with real-world applications in LLM key-value cache compression.

Open →

2026-06-11

How GitHub Apps Work — Build a Bot You Can @Mention

Interactive guide to GitHub Apps: app identity, JWTs, installation tokens, user OAuth, webhooks, and the full architecture for a bot that creates and reviews PRs when you @mention it on an issue.

Open →

2026-06-09

What Salesforce Learned from 20,000 Agent Deployments

Key lessons from Salesforce's analysis of 20,000 enterprise Agentforce deployments — architecture, failure patterns, and what it takes to succeed at scale.

Open →

2026-06-06

MacBook M4 Pro as a Local AI Powerhouse

How the MacBook Pro M4 Pro with 48GB unified memory can replace ChatGPT, Cursor, and Midjourney with fully local, private, offline AI tools.

Open →

2026-06-06

Running an AI-Native Engineering Org

An interactive learning guide to running an AI-native engineering organization — how planning, context gathering, code review, team composition, and metrics shift when agentic coding becomes the default.

Open →

2026-06-06

How Non-Engineers Learned to Ship with AI

How Sentry's growth team migrated 2,500 CMS pages to Git and unlocked AI-native workflows so non-technical contributors could ship like engineers.

Open →

2026-06-03

Skip-level One-on-ones: A Leader's Guide

An interactive learning guide to running effective skip-level one-on-ones — surface blind spots, transfer social risk, and build trust across organizational layers.

Open →

2026-06-01

The Private AI Business Blueprint

How to build a $1,000/mo private AI business for law firms using RTX 5090, Ollama, RAG, and Open WebUI — hardware, software, business models, and reality checks.

Open →

2026-05-30

How is Linear So Fast? — Interactive Breakdown

An interactive breakdown of how Linear achieves instant UI interactions — exploring sync engines, optimistic updates, and local-first architecture.

Open →

2026-05-30

Decoding LLM Model Names — Parameters, MoE, and Active Counts

Demystify AI model naming conventions — what parameter counts mean, how Mixture-of-Experts changes active vs total parameters, and why model size alone doesn't predict performance.

Open →

2026-05-30

Building Self-Improving Tax Agents with Codex

How to build tax agents that write, test, and improve their own code using OpenAI Codex — covering agent loops, self-evaluation, and iterative refinement patterns.

Open →