A hands-on workbook for practical generative AI

AI Workbench

ChatGPT, Claude & Practical Generative AI

Use AI to plan work, write and edit, create media, and build small automations. Every lab gives you fictional source material, a worked example, and a way to check the result.

Get the workbook · $10 Read the free sample

Digital PDF and practice files

600workbook pages

60original labs

4learning tracks

$10one-time price in USD

The learning path

Four ways to put AI to work.

LABS 01–15

Prompting and everyday work

Plan a week, trace claims to evidence, and compare assistants using the same work sample.

LABS 16–30

Business, writing and marketing

Write launch emails, plan campaigns, analyze feedback, and check spreadsheet formulas.

LABS 31–45

Images, video, voice and music

Direct images, video, voice, and music using production briefs, review criteria, and permission records.

LABS 46–60

Assistants, coding and automation

Test structured data, review code, and run an offline Python queue that checks drafts before dispatch.

Search the curriculum

Find a lab.

Search labs Learning track

60 labs

Plan a Week from Competing Commitments (pages 7–16)

Turn Meeting Notes into an Accountable Action Log (pages 17–26)

Build a Source-Locked Research Brief (pages 27–36)

Compare Two Assistants with a Blind Work Sample (pages 37–46)

Practice a Skill with an Adaptive Tutor (pages 47–56)

Turn an Ambiguous Email into a Decision Request (pages 57–66)

Rehearse a Difficult Conversation with a Roleplay (pages 67–76)

Clean a Small Dataset Without Losing the Raw Evidence (pages 77–86)

Extract Themes from Feedback Without Overcounting (pages 87–96)

Build a Reusable Prompt from a Failed First Draft (pages 97–106)

Synthesize Notes into a Decision Memo (pages 107–116)

Estimate a Project with Ranges and Dependencies (pages 117–126)

Verify a Claim Before It Enters a Presentation (pages 127–136)

Redact a Support Case Before Asking for Help (pages 137–146)

Hand Off a Repeatable Workflow with Stop Rules (pages 147–155)

Turn a Brief into a Launch Email (pages 156–165)

Edit a Draft Without Changing the Evidence (pages 166–175)

Build a Case Study from Interview Notes (pages 176–185)

Write Interface Copy for a Recoverable Error (pages 186–195)

Turn Brand Voice into a Reusable Brief (pages 196–205)

Plan a Campaign Around One Testable Question (pages 206–215)

Build a Search Brief from Reader Intent (pages 216–225)

Turn Service Options into a Clear Offer Page (pages 226–235)

Classify Customer Feedback Without Losing the Quotes (pages 236–245)

Prioritize a Pipeline with Explainable Rules (pages 246–255)

Clean an Order Export Before You Count Revenue (pages 256–265)

Total a Month from a Timestamped Ledger (pages 266–275)

Explain a Budget Variance from the Numbers (pages 276–285)

Model a Break-Even Point with Assumptions You Can Change (pages 286–295)

Build a Dashboard Rate That Survives Small Segments (pages 296–304)

Specify an image by its job (pages 305–314)

Edit a product photo with a change contract (pages 315–324)

Recompose a visual for three frames (pages 325–334)

Make a repeatable illustration series (pages 335–344)

Describe visual evidence with boundaries (pages 345–354)

Write alt text and a data equivalent (pages 355–364)

Budget a short video before generation (pages 365–374)

Control continuity between generated shots (pages 375–384)

Set limits for synthetic narration (pages 385–394)

Time narration with a spoken rehearsal (pages 395–404)

Make a caption track from final audio (pages 405–414)

Brief original music by function (pages 415–424)

Balance voice, music and useful sound (pages 425–434)

Adapt a video across languages (pages 435–444)

Release a media package with evidence (pages 445–453)

Design an assistant with a narrow job (pages 454–463)

Build a knowledge answer that cites its evidence (pages 464–473)

Defend against instructions inside reference text (pages 474–483)

Turn an API response into a checked record (pages 484–493)

Estimate request cost and build a stop rule (pages 494–503)

Build a tiny evaluation set (pages 504–513)

Put code review ahead of code generation (pages 514–523)

Trace a tool call before allowing a write (pages 524–533)

Create a dry run for a no-code workflow (pages 534–543)

Make retries safe (pages 544–552)

Bind approval to the reviewed draft (pages 553–562)

Choose between a rule, a model, and review (pages 563–572)

Log enough to debug without logging everything (pages 573–582)

Set a bounded stop rule for an agent loop (pages 583–591)

Capstone: Run an offline review queue (pages 592–600)

LAB 01

Plan a Week from Competing Commitments

A useful weekly plan begins by distinguishing commitments, estimates, and preferences. A commitment has a real time boundary or dependency. An estimate is a guess about duration and can be revised after work begins. A preference helps place flexible work without creating a new obligation. ChatGPT or Claude can help lay these out, but each proposed block still needs to fit inside the actual availability. Ask for a compact constraint table before asking for a schedule so that you can catch forgotten travel time, optimistic task estimates, or work that should move to a later week. Use a separate list of options for overflow. When the first schedule looks tidy, simulate a common interruption and check which blocks can move without creating new conflicts.

Workbook pages 7–16

Look inside

Read a real workbook page.

Browse one preview from each track, or open the 12-page PDF sample for a closer look at the format and opening exercises.

Preview a track

Open the 12-page sample

Workbook preview from Lab 1, Plan a Week from Competing Commitments — Lab 01 · Plan a Week from Competing Commitments · Page 10

Before you start

Common questions.

Where should a beginner start?

Start with the first track. It covers writing prompts, checking answers against sources, and comparing ChatGPT with Claude using the same exercise.

What do I need to run the exercises?

Use a chat assistant and a text editor for the prompt labs. Media labs can use your preferred image, video, or audio tool. The Python capstone runs offline and needs no API key. Some optional tools may require their own accounts.

What does the Python capstone do?

It processes synthetic support tickets, checks draft replies against a local catalog, holds uncertain cases for review, and prevents repeated simulated dispatch.

What format do I receive?

You receive the 600-page digital PDF and a ZIP with the workbook, practice files, and offline Python capstone.

The complete workbook

AI Workbench · $10

Get all 60 labs and keep the practice files for your own projects.

Open the Gumroad listing

USD base price. Applicable tax is shown at checkout.