Agentic Foundations: AI & AI Agents — From Chatbot to Operator

You already use AI. This module shows you the ceiling you've been using it under. In plain language, you'll learn what a large language model actually is, meet the two postures of Consultant Mode and Delegation Mode, and pick up the one-sentence filter the whole course runs on: the Work Off the Desk Test. Your lab is an audit of your own chat history, and you should expect a near-100% Consultant Mode result. That number is not a failure. It's the baseline the rest of the course exists to move.

In Module 1 you decided you need a hire. Before day one of the internship you need to read the résumé. The whole résumé is four slots: model, tools, memory, governance. In this module you'll strip the marketing off one real agent product with a single question, then spec your own first agent on half a page. You'll walk away holding two artifacts: a teardown verdict on a real product and your first agent spec.

Today you make your first hire. You'll install one agentic CLI tool, and either Claude Code or Codex will do. Then you'll verify it with an observable check and supervise it through three small tasks, approving every single action, exactly the way you'd watch a human intern on day one. You'll say no to it at least once, on purpose, and watch it stop. You walk away with a working agent on your machine and your first governance artifact: the day-one run log.

In Module 3 you hired one intern; today you draw the org chart for your one-person company's AI staff. It has four named roles: Mission Control, Router, Architect, and Worker, each with one job and a right-sized model. You'll learn the two laws that make the chart pay. The first is "Opus writes specs, Sonnet writes code." The second is that the split pays twice, because it makes the work cheaper AND easier to debug. Then you'll sketch one of your own repetitive tasks across the four roles and walk away holding a written verdict: agent-shaped or not. One thing before you start: you are not in the chart. You're the boss all four roles report to.

Right now, everything your Module 3 hire knows about how you work lives in one place: the tool you happen to be running this quarter. Today you fix that before it becomes expensive by splitting the brain (the rules and lessons you own, in plain files) from the body (whichever runtime executes them this quarter). You'll walk away holding your first two brain files: an AGENTS.md with the seven hard stops plus one of your own, and a WORKFLOW_RULES.md a stranger could follow cold, stored where they survive any platform change.

You hired the intern, drew the org chart, and wrote the employee handbook. Now comes the step a good manager never skips: writing the job description. You write it by watching the job get done first. For one full week you automate nothing and log every task you finish; then you distill that log into exactly eight prompts on one printed page. That page is the artifact Module 7's build depends on, and the discipline that produces it is the one almost everyone skips.

Everything you have built so far converges here: today, one of your eight prompts becomes your intern's first real assignment. You will spec it in the four slots, build it as a skill your CLI tool discovers and runs, supervise it at Trust Level 0 with every action approved and every run logged, and gate the output with the Work Off the Desk Test. Then comes the paperwork most managers skip. Write the promotion criteria into AGENTS.md before promotion is even possible. You walk away with a working skill and a paper trail a good manager could audit.

You made your first hire in Module 3 and gave it real work in Module 7. This final full module is workforce planning: you will run one real high-stakes prompt across two or more model families and record where they differ. Then, before the irresistibly convenient always-on product is offered to you, you will answer the Three Questions Before You Say Yes in writing about your setup as it exists today. You will walk away holding your first MODEL_NOTES.md entry, a completed THREE_QUESTIONS.md, and a declared exit ramp from this course.

The keynote ends by handing its audience a three-week assignment, and you have already done weeks one and two inside this course. This capstone turns the week after into a bounded management decision: document why the next prompt is safe and worthwhile, define its finished output, specify its four slots and governance boundaries, prove the start-date cadence, and write the verification plan. Pro learners submit the resulting SECOND_SKILL_DECLARATION.md against the /20 rubric, receive guided revision on attempt one even at a perfect score, and may pass on attempt two or later at 18/20 or above. Passing this ninth gate after the other eight earns L3 and completes the certificate path; free learners build and self-assess the same artifact without platform submission.