Start Here

Triadive is a field manual for the human / agent / robot triad — the operating environment that forms when long-running AI systems do real work on someone's behalf. It is built for operators who have decided to put more than a weekend into an agent setup, whether that setup is a single workspace, a small team, or a system that touches physical hardware.

This page is a short orientation. Five minutes here will save you an hour of wondering where to begin.

What this manual covers

The pieces on this site collect concepts, patterns, and lessons that keep coming back when an agent system has to last longer than a single chat session. They are written at the level of ideas, with OpenClaw used as the recurring worked example because that is where the practice is most visibly stabilizing — but the brand is broader than any one toolchain. The principles here are meant to outlive the platforms they came from.

The site deliberately does not chase news. It does not rank products. It does not run sponsored reviews. It is closer to a reference shelf than a feed.

How the manual is organized

Articles fall into five buckets. Reading order roughly tracks the buckets themselves.

• Field Notes — flagship essays. The longest pieces on the site, built around one durable question per essay. New operators often skip these on first pass and come back later; experienced operators tend to read them first.
• Concepts — short explainers for the vocabulary you will see everywhere else: agent loops, graphs, memory, prompting, orchestration. Read these to get the language.
• Workflows — pattern reviews. Each one names a pattern, names where it breaks, and names what to do instead. Useful when you are about to start something and want a sanity check.
• Lessons — anonymized, pattern-based reflections on what has gone wrong in real workspaces. The lessons are abstracted so they stay useful without revealing the specific operator, scale, or tooling that produced them.
• Glossary — short definitions of recurring terms, with links to the longer pieces that use them.

If you have five minutes, read the next section. If you have thirty, read the five pieces listed and skim the glossary.

Five pieces to read first

If you read nothing else on the site, read these. They are not the longest pieces; they are the ones that change how you read everything else.

1. Field Notes: The Three Amigos of the Agent Stack — a short essay on the human / agent / robot roles and why each role is load-bearing. The cleanest entry point on the site for a first-time visitor. 2. What Is an Agent Loop? — the operating rhythm of any agent system. Once the loop is visible, most of what the agent does becomes easier to read, debug, and improve. 3. Loops vs Graphs — the routing decision behind every non-trivial workflow. When does iteration become routing? When does a graph become overkill? 4. Memory: Short-Term, Long-Term, and Semantic — the three-layer memory model that lets an agent remember across sessions. The fourth layer (wiki memory) is in Wiki Memory. 5. Workflow Review: The Daily Check-in Loop — a concrete workflow pattern with named failure modes. Good template for designing any scheduled agent work.

How to read the buckets

A piece in Concepts is a definition with reasoning. It tells you what something is and why it earns its place. It does not tell you how to set one up.

A piece in Workflows is a pattern with trade-offs. It tells you what the pattern is good for, where it tends to break, and what to watch for when you adopt it.

A piece in Lessons is a story with the names removed. It tells you what went wrong, what the operator tried next, and what stuck.

A piece in Glossary is a definition with pointers. It tells you the shortest version of the term and links to the longer pieces that use it. Use these as a map, not as a destination.

A note on platform specificity

When a piece references OpenClaw surface area — sessions, cron, plugins, skills, memory layers — it is because that is the most visibly stable example of the pattern right now. When the pattern is a principle, the piece tries to write it so it can travel to other agent runtimes, other stacks, and other decades.

The site is not affiliated with any platform vendor. It does not accept payment for coverage. Editorial decisions are made on the durability of the principle, not the freshness of the release.

What to read next

Once the five pieces above are familiar, three more extend the picture in useful directions:

• Field Notes: Why we say "triad," not "stack" or "tools" — a short piece on vocabulary and what vocabulary commits you to.
• Lessons Learned: Memory Design Mistakes in Early Agent Work — the four most common memory failures new operators hit.
• Lessons Learned: Why Agents Stall Without Clear Termination — termination is the cheapest piece of agent design to skip and the most expensive to retrofit.

From there, the Concepts and Workflows indexes are the right next stop. They are organized by difficulty: beginner pieces first, intermediate second, advanced reserved for after the rest of the manual feels familiar.

Welcome.