Spec / 006 · curriculum
11 original lessons, in the order you should actually learn them. Model → prompt → agent → tools → memory → workflow → knowledge → agents together → evaluation → production. Building an agent that can perform a task is easy. Building one that can safely perform that task 10,000 times is the rest of this path.
These are not blog notes and not a live model run. Stage 10 (reliability) is inserted before advanced coding agents on purpose.
01
LLM Foundations
How large language models work, which one to pick, and what context and reasoning actually cost.
7 sections · 5 min · Aug 2026
What you learn
How LLMs work and which model to choose for a given job, budget, and risk.
Open lesson →02
Prompt Engineering
Templates are copy-paste roles. Engineering is how you reliably control behavior: system prompts, examples, seed crystals, and output contracts.
7 sections · 4 min · Aug 2026
What you learn
How to reliably control model behavior without pretending a gist is a program.
Open lesson →03
Agent Fundamentals
An agent is a system with a job, tools, state, and a stop condition — not a chat box with a costume.
6 sections · 3 min · Aug 2026
What you learn
How to build your first agent as an inspectable loop, not a longer prompt.
Open lesson →04
Tool-Using Agents
Tool calling is how models take actions. Tool design, memory, and middleware decide whether those actions are safe to repeat.
6 sections · 3 min · Aug 2026
What you learn
How agents interact with the real world through tools, memory, and a permissioned runtime.
Open lesson →05
MCP & Integrations
The Model Context Protocol is how agents connect to external capabilities without a new bespoke plugin for every vendor.
5 sections · 3 min · Aug 2026
What you learn
How to connect agents to external capabilities through a protocol instead of a pile of one-off plugins.
Open lesson →06
Agentic Workflows
Reliable multi-step systems are graphs with conditions, not a single while-loop that hopes.
6 sections · 3 min · Aug 2026
What you learn
How to build reliable multi-step systems with explicit control flow instead of an unbounded chat loop.
Open lesson →07
RAG & Knowledge
Agents work with private knowledge through retrieval, not by stuffing the company wiki into the system prompt.
6 sections · 3 min · Aug 2026
What you learn
How agents work with private knowledge without pretending the model memorized your corpus.
Open lesson →08
Multi-Agent Systems
Multi-agent systems are workflows with roles, handoffs, and shared state — not extra names on the same chat.
6 sections · 3 min · Aug 2026
What you learn
How to coordinate multiple specialized agents with contracts, not theater.
Open lesson →09
Production Engineering
An agent that works in a notebook is not a product. Tracing, evals, local loops, and deployment are the job.
6 sections · 3 min · Aug 2026
What you learn
How to make agents reliable enough to ship: traces, evals, and a deploy path with rollback.
Open lesson →10
Reliability Engineering
Building an agent that can perform a task is easy. Building one that can safely perform that task 10,000 times is the actual job.
7 sections · 3 min · Aug 2026
What you learn
How to keep agents safe, bounded, and repeatable in production — not only clever in a demo.
Open lesson →11
Advanced Agents
Modern autonomous systems edit files, run in sandboxes, see images, and sometimes write other agents — still behind budgets and gates.
6 sections · 3 min · Aug 2026
What you learn
How modern autonomous AI systems are built — and why they still need the previous ten lessons.
Open lesson →If you are building Hermes- or OpenClaw-style autonomous agents, master Tier 1 before you spawn subagents.
Tier 1 · Must know
Tier 2 · Production skills
Tier 3 · Advanced
Labs and a shorter on-ramp live on /learn.