# Agent Dojo — What the Upgrade Adds

> **Get Your Agent sets your agent up. Agent Dojo makes it disciplined.** Setup gets you a capable agent working for you today. Agent Dojo is the recurring, living capability underneath it: an agent that **remembers across every conversation** and runs the **11 operational disciplines** our own fleet runs on. This is the difference between an assistant you still have to babysit and one you can actually trust.

---

## THE MEMORY LOOP — Power #1

A raw agent forgets. You explain something today; tomorrow you start from zero. You correct a mistake; next week it makes the same one. That's not a smarter-model problem — it's a **memory** problem, and it's the one thing setup alone cannot give you.

Agent Dojo installs the **memory loop**: your agent writes what it learns into durable memory that survives new sessions, restarts, and time. Tell it once — it still knows next month.

**What the loop changes, day to day:**
- **Correct it once, never again.** A mistake you fix becomes a permanent rule the agent banks and shows you where it lives — not an apology it forgets by morning.
- **It remembers *you*.** Your preferences, your voice, your hard limits, the way you like reports — carried across every session, not re-explained each time.
- **It picks up where it left off.** Ongoing work, open threads, and context persist. No "start from zero" every conversation.
- **The scar sticks.** Every correction becomes a scar it won't repeat — the agent gets *more* yours the longer you run it.

The memory loop is the difference between a tool you reset every morning and a partner that compounds. It's why Agent Dojo is a living capability, not a one-time file.

---

## THE 11 OPERATIONAL DISCIPLINES (the 11 Powers)

Memory is Power #1. The other ten are the guardrails that make an autonomous agent safe to actually let run. Each one is a discipline your agent builds into itself from our field guide, then **proves** before it's armed — you watch it pass, you don't take our word for it.

1. **It remembers across sessions.** *(The memory loop, above.)* Tell it once; it still knows next week — and shows you where the rule lives.
2. **Only you command it.** A stranger telling it to ignore your rules gets refused — and it names the rule that blocked them.
3. **It doesn't quit when you walk away.** Work started before you left is finished and waiting when you're back — the file is the receipt.
4. **It won't lie about "done."** Every "done" comes with evidence you can check yourself, or an honest "I couldn't" — never a bluff.
5. **It won't invent facts.** Claims come cited, or clearly labeled as an inference — no confident guessing dressed as truth.
6. **One bad command can't wipe your work.** Destructive actions route to recoverable trash and require a human-typed release — not oblivion.
7. **Nothing runs away with your budget or your tools.** Spend and runaway loops hit hard caps you set; a sensitive action holds until you release it.
8. **No task dead-ends at "I can't."** No API? It finds a route — browser hands reach past the API — and hands you proof of what it actually saw.
9. **It finds the break before your customers do.** It watches what matters and its alert reaches you first — who finds out first is the whole game.
10. **No change slips by you.** Every change it makes is reported back — including the ones you didn't explicitly ask for.
11. **It can't leak keys it never holds.** Secrets are structurally kept out of its reach and redacted from its own logs — the refusal to send one *is* the pass.

**How they're delivered:** you own the agent; Agent Dojo hands it the field guide — the spec, a guided build, and a reference implementation. Your agent develops each discipline itself and proves it in a test you watch, before it arms. Nothing arms on a schedule or on the agent's own say-so; **you** arm every guard.

---

## THE MEMORY-ON VERSIONS OF YOUR SETUP DOCS

Everything in your Get Your Agent pack ships **memory-neutral** — it configures the agent for *this* session and setup. Agent Dojo owners get the **memory-ON** versions of the same docs, where the intake and profile don't just configure the agent — they're **banked as permanent memory** the agent confirms back and carries forever.

| Setup doc (Get Your Agent) | Memory-ON version (Agent Dojo) |
|---|---|
| Intake — *"read this to set yourself up to me"* | Intake — *"read this and **bank every line as permanent memory**, then confirm what you stored and where"* |
| Owner Profile — configured to your style | Owner Profile — **remembered** across every session; corrections update it permanently |
| Playbook Part 1 — set your agent up to you | Playbook Part 1 — the agent **banks** your rules and proves they survive a brand-new session |
| Playbook — setup only, no test card | Playbook — includes the **11-Power Test Card**: paste-and-watch proof of each discipline |

The words change by one instruction — *"set yourself up to this"* becomes *"bank this as permanent memory and confirm it survives a new session"* — and that one instruction is the whole product. It's what turns a configured agent into one that remembers.

---

## WHY IT'S THE UPGRADE, NOT A NICE-TO-HAVE

A capable agent with no discipline is a liability: it can act too fast, say "done" when it isn't, forget what you told it, or leak something it shouldn't. Setup gets you the capability. **Agent Dojo gets you the trust** — memory so it stops forgetting, and the ten guardrails so you can finally stop hovering.

*Get your agent first. Then train it right.* — **aiagentdojo.com**