THG AI Usage & Prompting
THG Learn · Track ai-usage · whole company

AI Usage & Prompting

Ten modules. Each one teaches one thing, makes you play it, then makes you prove it. Brutally honest, THG‑specific, and about two hours end to end. Do them in order; each one assumes the last.

Not startedStart with Module 01. Ten minutes.
Start Module 01 →
WHY THIS TRACK EXISTSEveryone at Tom Horn Gaming now works next to agents that can edit code, send messages and touch production. Most of us were never taught how. This is that teaching. It is not optional, and it is not long.

How it works

orderTen modules, in sequenceEach module unlocks the next. The capstone needs everything before it.
formatTruth → teach → play → proveA brutal one‑liner, a short lesson, a playable game with real THG cases, a five‑question quiz.
pass mark80% on every quizFour of five. Fail and you retake it on the spot. Only passes are recorded.
recordProgress is saved to your nameBehind Zero Trust, automatically. You will see "Progress saved" after each pass. If you see "Not saved", press Retry.

Modules

01

Terminology, no fluff

Model, agent, subagent, context. Say them right or blame the wrong thing forever.

available~10 min · drag game · quickfire
02

What a lousy prompt costs

Vague in, wrong out, hour gone. A live grader shows you the bill for your own prompt.

available~12 min · live prompt grader
03

Anatomy of a good prompt

Role, context, constraints, output format. Build one from blocks and watch the result change.

available~12 min · block builder
04

Reverse‑prompting & virtual colleagues

Ask for the questions first. Then borrow John, Charlie, Maya, Leo or Sophie's lens.

available~12 min · pick‑the‑colleague
05

Subagents & second opinions

When to spawn fresh eyes, when to cross‑check two models, and when not to bother.

available~12 min · subagent drill
06

Context hygiene

Never build and review on the same desk. Watch a session bloat until it lies to you.

available~10 min · bloat simulator
07

AI safety: the red lines

What never goes into an AI tool: secrets, player data, confidential deals. Classify or crack the company.

available~12 min · safe‑or‑not classifier
08

When the AI "doesn't work"

Dead token, drift, "stupid" questions, crashed subagent, escalation. Six situations, six fixes, no guessing.

available~12 min · branching stuck cases
09

Failure theatre

Six real‑shaped disasters, staged. Make the call, watch the terminal, then watch the bill.

available~15 min · damage meter
10

Capstone: ship it or crack it

One real task, one agent, every trap in order. Score 80 to win. Red lines end the run.

available~20 min · score‑gated mission
In‑house only · Tom Horn Gaming · deployed behind Zero Trust at ai-usage.iguanatech.sk