Course 1 — Master Course

Harness
Engineering

Twelve modules on building production AI agent harnesses. The execution loop, tool design, context management, memory, sandboxing, permissions, error handling, verification, observability, and security. The thesis: the model is 1.6% of an agent. The harness is the other 98.4%. Twenty-one deep-dives — from Pi (the minimal baseline) to Tau (the educational reference) to LangGraph, CrewAI, Mastra, and the full coding-agent ecosystem.

~26.5

hours

12

modules

21

deep-dives

Deep-Dives

13 studies

Pi: The Minimal Baseline

What you get with 4 tools and a <1,000-token system prompt. Pi is the reference thin harness — every other deep-dive in the curriculum compares against it.

Hermes: Layered Persistent Memory

Self-evolving skills. The depth play. The memory reference. 33,000+ stars. Overtook OpenClaw on OpenRouter May 10, 2026 (224B tokens/day). On the breadth-vs-depth split, unambiguously the depth competitor.

NemoClaw: The Governance-Focused Harness

NVIDIA's hardened OpenClaw fork. NeMo Guardrails. OpenShell sandboxes. Policy enforced OUTSIDE the agent's reach. The governance reference. The anti-Tau. 39/60 (+4 over OpenClaw) — the +4 comes entirely from the boundary, not the agent.

LangGraph: Graph-Based State Machines

The most architecturally distinct harness. Explicit nodes and edges. Super-step checkpoints. interrupt() for HITL. Used INSIDE Claude Code. The orchestration baseline and the Module 8 reference. 37/60 — maxes on Modules 1 (Loop) and 8 (State), both 5/5.

OpenAI Agents SDK: 2-Layer Harness/Compute Split

The most architecturally thoughtful SDK. 2-layer harness/compute split. 7 sandbox providers. Handoffs + agents-as-tools. The SDK reference. Module 5 (Sandboxing): 5/5 — the highest-scoring sandbox in the roster.

CrewAI: Role-Based Multi-Agent (Emergent Coordination)

45,900+ stars. Most accessible multi-agent framework. Role-based sequential/concurrent crews. 1.8s avg latency. The emergent-coordination pole — paired with LangGraph (DD-10) as the declared-coordination pole, both load-bearing for Course 4 E09 (Multi-Agent Orchestration).

OpenHarness: Academic Baseline (Inspectability as Product)

HKUDS lab, April 2026. Research reproducibility, not production. Auto-compaction as a designed four-part mechanism. The harness Module 3 points at when it teaches compaction. 30/60 — the score reflects what it is not (production), not a failure of what it is (the cleanest mechanism reference in the roster).

Mastra: Observability Primitives (Native Pattern + Read/Write Memory)

TypeScript-first SDK. Observability as a first-class primitive (native event emission, not a wrapper). Explicit read/write memory tiers. The Module 10 reference (5/5) and the Module 4.3 reference (5/5). The only roster harness where the memory-poisoning defense is structural rather than advisory. 34/60 — the observability-and-memory shape.

Command Code: The Taste-Learning Harness

3,500+ stars. $5M seed. Closed-source TypeScript. The first harness that learns your coding taste from accept/reject signals and persists it as Markdown. A fundamentally new feedback model.

ZeroClaw: The Rust Microkernel Harness

32,000+ stars. Apache-2.0. Rust single-binary. Trait-driven microkernel with 6-layer safety. 25+ LLM providers. The thin-by-philosophy, medium-by-implementation answer to OpenClaw.

PicoClaw: The Edge-Hardware Harness

30,000+ stars. MIT. Go single-binary. Cross-compiles to RISC-V, MIPS, LoongArch. Real I2C/SPI/Serial tools for Sipeed hardware. The only harness that reads sensors and drives peripherals.

MCP: Building Production Tool Servers

The wire protocol that turns hardcoded tool registries into discoverable, versioned, independently-deployable services. The platform turn. JSON-RPC 2.0, stdio/HTTP+SSE, tools/resources/prompts, capability negotiation, JSON Schema validation, registry/discovery, production failure modes. With a real TypeScript MCP server.

Cost-Aware Agents: Budget as a First-Class Constraint

Why agents cost 1,000x more than chat, and how to add budget enforcement to any agent loop. The @token_meter interceptor pattern: throttle at 80%, rollback at 100%, circuit-break on retry storms. Prompt caching (90% discount). The prime directive. With a real TypeScript cost-aware loop wrapper.

Prerequisite: None. This is the foundational course. Status: complete.