Clear Promise
Each lesson teaches one practical move and shows what the matching paid class helps you build next.
Preview Reader
Use one practical sample before choosing a Build.
These are the browser-readable The Practical AI Workbench preview lessons. Each one teaches a small useful move, names the paid artifact, and gives a stop point if the free lesson solves enough for now.
Preview Clarity
Use the reader like a small test drive: one sample, one low-risk example, one decision about whether the paid artifact path is worth buying.
Who This Is For
Readers comparing practical AI training paths without needing enterprise jargon, vendor lock-in, or a broad tool tour.
First 15 Minutes
Pick the preview that matches your current bottleneck, try it on safe material, and write what the full class would help you finish.
Quality Checks
The preview is working if you can name the useful move, paid artifact, safety boundary, tool-agnostic transfer, and next step.
Choose A Preview
Start with the friction you actually have today. Prompt Foundations and Make Your Work Legible are the strongest first tests for most self-starters.
Prompt Foundations
AI output is vague, generic, or hard to review.
Make Your Work Legible
Files, notes, screenshots, or source material are too messy to brief AI well.
AI-Ready Work OS
You keep re-explaining the same work, sources, and boundaries.
AI Surfaces
You do not know whether work belongs in chat, a project, a harness, an agent, a runtime, or human review.
Basic Agents
One recurring helper idea might need a narrow, reviewable agent.
Agentic Workflows
A recurring process needs gates, evidence, evals, and receipts.
Prompt Foundations Preview
Turn a weak prompt into a work brief.
A good prompt is not magic wording. It is a short work brief: role, context, task, constraints, output format, and review criteria.
Starting Prompt
Help me write a follow-up email after a workshop.What Is Missing
| Missing Piece | Why It Matters |
|---|---|
| Audience | The email should sound different depending on who attended. |
| Workshop artifact | The email should remind participants what they built. |
| Next action | Follow-up should point to one concrete step. |
| Tone | "Sound good" often becomes generic or promotional. |
| Review criteria | You need a way to judge whether the draft is usable. |
Stronger Prompt
You are helping me draft a workshop follow-up email.
Audience:
Self-directed professionals who attended a 90-minute workshop on cleaning messy work material for better AI use.
Context:
Each participant started a context packet for one real project.
Task:
Write a concise follow-up email that thanks participants, reinforces the main idea, and gives one next action for the next seven days.
Tone:
Warm, plain, practical, not salesy.
Avoid:
Hype, guaranteed outcomes, and long paragraphs.
Return:
Subject line, email body under 150 words, and three review questions before sending.Try This
Rewrite one prompt you have used recently. Add audience, purpose, source material, output format, and review criteria.
Bridge To The Full Build
Free Win
You leave with one stronger prompt brief and a clearer sense of why the original request was too vague.
Paid Build Unlocks
A reusable prompt pack for briefing, reviewing, revising, researching, meeting follow-ups, and recurring AI-assisted work.
Best Fit If
Your AI outputs sound polished but generic, or you keep rewriting prompts from scratch because the request never names the real work.
Make Your Work Legible Preview
Clean a folder without losing control.
AI can help inspect a folder listing, but it should not guess file contents, delete files, overwrite names, or handle sensitive material casually.
Sample Listing
notes final.docx
new notes.docx
client thing.pdf
Screenshot 2026-06-01 at 8.41.12 PM.png
draft2.md
old plan maybe.txt
IMG_4188.jpeg
contract.pdfSafe Cleanup Rules
- Work from a read-only listing.
- Do not delete anything.
- Do not overwrite anything.
- Do not assume file contents from names alone.
- Route sensitive or ambiguous items to Review.
- Save a cleanup receipt before applying changes.
Dry-Run Prompt Starter
You are helping me clean a folder so the contents become usable work context.
You can only see the listing I provide. Do not claim to know file contents unless I provide them.
Return:
Original name | Proposed name | Suggested bucket | Source status | Reason | Confidence | Human check needed
Rules:
Never delete, never overwrite, preserve extensions, and route ambiguity to Review.Try This
Pick one folder with 10-40 items. Create a read-only inventory before asking AI for cleanup help.
Bridge To The Full Build
Free Win
You create a safer read-only inventory and dry-run cleanup table before changing names, moving files, or asking AI to reason from messy source material.
Paid Build Unlocks
A repeatable data-cleaning and context-packet workflow for files, notes, screenshots, PDFs, drafts, source-of-truth checks, and cleanup receipts.
Best Fit If
Your AI work is unreliable because the source material is scattered, stale, ambiguous, or too messy to hand off confidently.
AI-Ready Work OS Preview
Build the first three Work OS files.
A Work OS is not a giant autonomous system. It is a readable operating layer for AI-assisted work: who you are, what is active, what sources matter, what rules apply, and what receipt should be left.
Starter Folder
work-os/
PERSONAL_CONTEXT.md
ACTIVE_WORK.md
SOURCE_MAP.mdPERSONAL_CONTEXT.md
Your role, current focus, output preferences, and approval boundaries.
ACTIVE_WORK.md
The project status, next action, and source of truth for active work.
SOURCE_MAP.md
The places an assistant may trust, inspect, ignore, or ask about.
Try This
Create only these three files. Then ask an AI assistant to summarize your active work using them. If the answer is generic, improve the source map and active work notes.
Bridge To The Full Build
Free Win
You start a tiny Work OS with three files that reduce re-explaining and make the next AI handoff easier to inspect.
Paid Build Unlocks
A one-lane operating layer with context files, source rules, AGENTS.md and CLAUDE.md starter notes, receipts, keeping-current routines, and a handoff test.
Best Fit If
You use AI repeatedly on active work and keep losing the project state, source boundaries, decisions, or handoff history.
AI Surfaces Preview
Choose the right AI surface.
The durable skill is not picking a favorite tool. It is routing work to the right surface.
Surface Map
| Surface | Use When |
|---|---|
| Chat | One-off drafting, critique, explanation, or brainstorming. |
| Project | Work benefits from saved context, examples, and recurring standards. |
| Harness | Work needs local files, edits, terminal checks, or verification. |
| Always-on agent candidate | Work repeats and needs monitoring, but should start read-only. |
| Runtime | Workflow is stable, tested, scheduled, and gated. |
| Human-only | Judgment, approval, sensitive context, or accountability is central. |
Try This
List five real tasks. For each, decide whether it needs chat, project, harness, agent, runtime, or human-only handling.
Bridge To The Full Build
Free Win
You route five real tasks to the right surface instead of forcing every request into the same chat window.
Paid Build Unlocks
A platform-agnostic routing map across ChatGPT, Claude, Gemini, projects, Codex, Claude Code, Antigravity, CLIs, MCP, connectors, runtimes, and human review.
Best Fit If
You are tool-aware but still unsure where work belongs, when a harness matters, or when human-only handling is the more professional choice.
Basic Agents Preview
Decide: agent or not agent?
A first agent should be narrow, tested, gated, and useful. If the task does not repeat or cannot be reviewed, it probably should stay a prompt or workflow note.
Example Task
Prepare a weekly project status draft from active work notes.
Agent Decision
| Question | Answer |
|---|---|
| Does it repeat? | Yes, weekly. |
| Does it need stable context? | Yes, active work notes and prior receipts. |
| Is the output reviewable? | Yes, a human can inspect the draft. |
| Are permissions narrow? | Yes, read and draft only. |
| Should it send updates? | No. Human approval required. |
Launch Boundary
Do not send, publish, edit files, archive work, mark tasks complete, or change priorities.Try This
Choose one recurring helper idea. Write three non-goals before writing the agent prompt.
Bridge To The Full Build
Free Win
You decide whether one helper idea is actually agent-shaped and name the non-goals that keep it bounded.
Paid Build Unlocks
A narrow first-agent pack with charter, context packet, permissions, loop, stop rules, tests, escalation rules, and receipts.
Best Fit If
You have recurring work that feels automatable, but you want an agent that drafts, checks, and escalates before it ever acts too broadly.
Agentic Workflows Preview
Turn a recurring task into a workflow candidate.
An agentic workflow is not a long prompt. It is a staged process with trigger, inputs, actions, human gates, outputs, evals, and receipts.
Candidate
Recurring task: weekly operating brief.
Trigger: Friday afternoon or before weekly planning.
Output: brief with priorities, blockers, decisions needed, next actions, and receipt.
Use-Case Score
| Criterion | Score | Note |
|---|---|---|
| Frequency | 5 | Weekly. |
| Friction | 4 | Easy to skip. |
| Value | 5 | Improves planning. |
| Context readiness | 3 | Needs reliable notes. |
| Review clarity | 4 | Human can inspect. |
| Risk manageability | 5 | Internal and draft-only. |
Try This
Pick one recurring task and write its trigger, inputs, output, and human gate before asking AI to help with it.
Bridge To The Full Build
Free Win
You turn one recurring task into a workflow candidate with trigger, inputs, output, and human gate before automation enters the picture.
Paid Build Unlocks
A repeatable human-plus-agent workflow with staged actions, context handoffs, evidence, failure handling, evals, receipts, and SOP.
Best Fit If
You want repeatable AI-supported work that stays inspectable, maintainable, and human-reviewed instead of becoming a fragile mega-prompt.
After One Preview
A preview is doing its job if you can name the useful move, the paid artifact, the safety boundary, and the next step without someone explaining it live.
Free Preview Solved It?
Stop there. The preview is allowed to be enough if it helped you do the small move you needed.
Choosing A Class?
Compare the free move to the paid artifact path before buying.
Ready For Access?
Use the self-serve checkout page for the matching Build, Starter Bundle, or Founding Lifetime Access.
Choose Your Next Step
After one preview, decide whether to stop, buy one Build, use the Starter Bundle, or choose Founding Lifetime Access. The right next step is the smallest one that matches the artifact you actually need.
Stop Here
Use the free move and come back later if the same bottleneck keeps showing up.
Buy One Build
Choose the matching $49 Build when you want the full student path, templates, examples, practice labs, capstone, rubric, and checkpoint.
Use Starter Bundle
Choose the $89 bundle when weak prompts and messy source material are both part of the problem.
Choose Lifetime Access
Choose the $199 founding lifetime path when you want all six Builds and the connected artifact chain.
Safety boundary: do not use private, regulated, client, student, patient, employee, legal, financial, or medical material in a public preview. Review AI output before using it.