# Capstone — The Agentic AI Manifesto

**Released after:** Session 3 · drafted across the build period
**Presented at:** Session 4 · Showcase
**Length:** One page. **Required for certification.**

## Why this matters

Everything else in the course was preparation for this. The manifesto is where you turn a
week of mechanics, capabilities, and caveats into a position your organization can act on.
It's deliberately short — one page forces clarity. You'll present it in the Session 4
showcase and leave with a document you can actually circulate.

## The format

A one-page declaration with three parts. Write it for *your* organization, in *your* voice.

### 1. Scope — where the agent goes to work

Name **one** concrete place you will put an AI agent or AI-assisted workflow to work in
the **next 90 days**. Be specific enough that someone could start Monday:

- The function and the task.
- Why it's the right first move (high value, low risk, mature tooling).
- What "working" looks like — the outcome you'll measure, not the activity.

### 2. Insight — what changed how you'll lead

State the **one idea** from this week that changed how you think about AI. Draw on the
concrete material, not the headlines. For example:

- "Because the context window is finite and the middle is unreliable, we will design AI
  workflows around small, well-scoped tasks — not one giant always-on assistant."
- "Because a raw model has no side effects, our risk lives entirely at the boundary — so
  that's where our controls go."
- "Because a skill adds context and an agent removes it, our first investment is *skills*:
  writing down how we already work, before automating anything."

### 3. Governance — the guardrails you insist on

The non-negotiables you will require from day one. Reference the tiered framework from
Session 4. At minimum:

- Which tier this first use case falls in, and what oversight that demands.
- The data and IP rule (what may never be sent to a third-party model).
- Who owns the outcome — because the AI can't be accountable, a human must be.

## Deliverable & showcase

- A **one-page** document (PDF or doc). Longer is not better.
- A **3-minute** presentation in Session 4: Scope, Insight, Governance — one minute each.
- Be ready for one question from the cohort.

## Assessment — what earns certification

| Dimension | What we look for |
|---|---|
| **Concrete** | A real, named, 90-day-startable use case — not "explore AI." |
| **Grounded** | The insight reflects how these systems actually work, learned this week. |
| **Responsible** | Governance is specific and matched to the use case's risk tier. |
| **Owned** | A named human is accountable for the outcome. |

## A note on using AI to write it

Use Claude to draft, sharpen, and pressure-test your manifesto — that's entirely in the
spirit of the course. But the **position must be yours**. In the showcase you'll be asked
*why* — and "the AI suggested it" is not an answer that ships.