Deep dives

Multi-part series that build one topic from first principles to production.

Search ⌘K

Deep dives

Five paths, each with a job.

Choose the path that matches what you are trying to build or debug. The cards lead with the reason to read, then show the first few steps in the path.

Choose by need

Framework internals Trace the runtime Harness strategy Map the harness Agent reliability Harden an agent AI engineering Learn the stack Streaming systems Trace the stream

Index

Choose the next thing to read

Framework internals

Harness Internals

Read Flue as a source-pinned runtime: sessions, provider seams, compaction, tools, run APIs, and deployment targets.

Best when

You want to understand where an agent framework owns behavior and where it rents the model loop.

You leave with

A practical map for debugging or extending Flue without blurring runtime boundaries.

8-part path Senior about 202 min straight through
Starts with 1 Runtime Map 2 Session Tree, Leaf, And Replay Safety 3 The Pi-ai Seam
Senior Agent Harness Runtime Boundaries Flue
Trace the runtime

Harness strategy

Harness Engineering: The Compounding Stack

Build the layer around the model: context, tools, memory, evaluation, and the habits that make it improve.

Best when

You are past prompts and need a platform that compounds across real engineering work.

You leave with

A map for turning scattered agent tricks into an operating system for teams.

13-part path Senior to Advanced about 296 min straight through
Starts with 1 Harness Engineering — What This Series Is, and Why You Should Read It in Order 2 What a Harness Actually Is (and What It Is Not) 3 The Four Primitives Every Working Agent System Has
SeniorAdvanced Harness Engineering Four Primitives Compounding Moat
Map the harness

Agent reliability

Production Agents Deep Dive

Take the demo loop apart and add retries, state, cost control, human review, durability, security, and tests.

Best when

You are putting agents near users, money, private data, or irreversible actions.

You leave with

A checklist for the parts that fail after launch.

9-part path Senior about 205 min straight through
Starts with 1 Production Agents Overview - The Loop Is 20% of the Work 2 Idempotency & Safe Retries - The Stripe Pattern for Agents 3 State Persistence & Agent Memory - The Complete Domain
Senior Agent Architecture Production Requirements Failure Modes
Harden an agent

AI engineering

AI Engineering Fundamentals

Connect tokens, embeddings, attention, retrieval, agents, and evals into one working mental model.

Best when

You want the AI stack to feel connected instead of like a list of borrowed terms.

You leave with

A grounded way to debug LLM systems without hand-waving.

8-part path Intermediate about 108 min straight through
Starts with 1 Text to Tokens - The Foundation 2 Tokens to Embeddings - Vectors That Capture Meaning 3 Embeddings to Attention - Relating Tokens to Each Other
Intermediate What Is A Model Probability Basics Math Intuitions
Learn the stack

Streaming systems

Kafka Deep Dive

Trace brokers, producers, consumers, logs, transactions, and event-sourced designs as one system.

Best when

You debug event streams, offset drift, ordering bugs, or exactly-once claims.

You leave with

A practical model of Kafka failure, recovery, and tradeoffs.

6-part path Senior to Advanced about 208 min straight through
Starts with 1 Kafka Architecture - Core Concepts 2 Producer Mechanics - Under the Hood 3 Consumer Groups and Rebalancing
SeniorAdvanced Topic Partitioning Consumer Groups Log Based Storage
Trace the stream