Become an AI expert.Become anAI expert.

The four things that turn a useless first draft into a usable one. The argument of the section: prompting is not a bag of tricks, it is a supply problem. You get generic answers because you handed over a generic situation.

1. 01Vague in, vague out example
2. 02The four ingredients
3. 03Role and task
4. 04Context
5. 05Format
6. 06Example: the same request, twice example
7. 07Show, do not just describe
8. 08Try it: rewrite one of your own try it
9. 09Quiz quiz

Where section 07 was about supplying information, this is about controlling the process. The order of the thinking, the order of the work, the shape of what comes back, and how to revise a draft without losing the good parts: the moves that matter once a job takes more than one exchange.

1. 01Ask for the thinking first
2. 02Break long jobs into steps
3. 03Chaining, and where to stand
4. 04Ask for a fixed shape when something else reads it
5. 05Name the fields before you ask
6. 06Make it criticise itself
7. 07Fixing a draft without losing the good parts
8. 08Try it: build something you will reuse try it
9. 09Quiz quiz

Every answer you have ever received was shaped by instructions you did not write. This is what those instructions are, who sets them, what happens when they disagree with yours, why a model seems to forget them, and why an instruction steers the model but cannot enforce anything.

1. 01There is always a prompt you did not write overview
2. 02The layers
3. 03What is actually in one
4. 04Where your own instructions actually go
5. 05Example: one question, three system prompts example
6. 06Writing one that works
7. 07What actually hurts
8. 08When the layers disagree
9. 09Text it fetched is not text that instructs extended reading
10. 10Why it seems to forget
11. 11It is not a secret
12. 12Try it: write one for a job you do try it
13. 13Quiz quiz

Fifteen minutes of setup that pays back on every conversation. Four ways a tool can remember things about you, what each is for, how memory works underneath, and the ways it goes wrong without telling you.

1. 01Stop explaining yourself every time overview
2. 02Four different kinds of remembering
3. 03What belongs in standing instructions
4. 04What does not belong in them
5. 05Projects and workspaces
6. 06How memory actually works
7. 07What gets saved, and how to see it
8. 08The quiet failure extended reading
9. 09Memory when other people are involved
10. 10Try it: fifteen minutes, paid back daily try it
11. 11Quiz quiz

How an assistant gets given a whole procedure instead of a message, and how it gets plugged into the systems you already use. Both are genuinely useful, both hand something real away, and both are routinely installed by people who have never been told what they are agreeing to.

1. 01Handing over more than a message overview
2. 02What a skill is
3. 03How it loads, and why that matters
4. 04Example: one skill, start to finish example
5. 05Risk one: instructions are trusted
6. 06Risk two: where it came from
7. 07Risk three: the quiet ones
8. 08Reading one before you use it
9. 09Connectors and MCP
10. 10What connecting one actually grants
11. 11The risks specific to connections
12. 12Deciding to connect one
13. 13Try it: audit what you have already connected try it
14. 14Quiz quiz

The other half of everyday use. Reading a photo, making a picture, talking instead of typing, getting a first draft of a deck out of a document. The point is not that these features exist. It is the handful of things each one is good at, and the specific places each one lets you down.

1. 01A photo is a prompt
2. 02What it reads well, and what it gets wrong
3. 03Cameras that watch, count and recognise
4. 04How a picture gets made
5. 05What that mechanism explains
6. 06Making pictures, honestly
7. 07Where making pictures still falls down
8. 08Editing rather than starting again
9. 09Talking instead of typing
10. 10Transcription and its ceiling
11. 11Opening doors, and who it fails
12. 12Documents, slides and spreadsheets
13. 13Where this quietly changes the job
14. 14Quiz quiz

Where it saves hours in an ordinary week: drafts, meetings, inboxes and thinking out loud. This is the turn from what it can do to what you do with it, and the audit at the end is what makes the final plan in section 25 possible.

1. 01Two questions before you hand anything over
2. 02Example: four tasks, worked through example
3. 03Draft, do not publish extended reading
4. 04Meetings
5. 05Recording other people
6. 06Inboxes
7. 07Thinking out loud
8. 08Try it: audit your week try it
9. 09Quiz quiz

Looking things up, long documents, and numbers, with the specific ways each one fails. This is the section that turns “the AI got it wrong” into “I asked it to do something it structurally cannot do”.

1. 01Give it the document
2. 02What is actually happening when you “ask your documents”
3. 03How the looking-up goes wrong
4. 04The thing that is not there extended reading
5. 05Long documents
6. 06Map first, then drill
7. 07Numbers deserve a calculator
8. 08Did it calculate, or did it guess?
9. 09Asking for sources, and meaning it
10. 10Quiz quiz

The “no code needed” promise, actually delivered. Having AI build the thing that does the job, whether a formula, a workflow or a small script, rather than doing the job for you each time.

1. 01Do the job, or build the thing that does the job overview
2. 02Is it worth automating at all?
3. 03Writing formulas you could not write yourself
4. 04Test it on rows where you know the answer extended reading
5. 05Workflow builders
6. 06Four things to check before you switch it on
7. 07Scripts and scheduled tasks
8. 08You own it whether or not you can read it extended reading
9. 09Try it: automate one thing try it
10. 10Quiz quiz

What changes when a model can take actions rather than only answer. Section 15 was AI helping you build automation; this is AI being the automation. The question of whether an action can be undone starts here and carries through the rest of the course.

1. 01From answering to doing
2. 02Where the line actually goes
3. 03Example: the line, drawn example
4. 04Instructions from strangers
5. 05Walkthrough: an agent that reads your email example
6. 06What “connected” actually grants
7. 07Read-only is a different product extended reading
8. 08It aims at what you measure extended reading
9. 09When agents hand work to other agents
10. 10When your agent meets theirs
11. 11Where agents genuinely pay off
12. 12Quiz quiz

The failures that cost people their reputation and their money, and the habits that catch them. Every earlier example that went wrong is classified here, each kind with its own check, and the section ends with the people who will try to fool you on purpose.

1. 01Confidence tells you nothing extended reading
2. 02Five ways of being wrong
3. 03The one that is hardest to catch extended reading
4. 04The things to check every single time
5. 05Bias it absorbed
6. 06The bias in you
7. 07It agrees with you too much
8. 08Asking it why is not the same as knowing why
9. 09Untrusted text, one more time
10. 10When the voice on the call is not them
11. 11Checking a person is who they seem
12. 12Try it: find the error try it
13. 13Quiz quiz

What you can paste in, what you cannot, and how to write a rule people actually follow. Section 03 covered where the machines physically are; this is about what happens to your words once they get there.

1. 01Where do your words actually go?
2. 02Four questions worth an actual answer
3. 03The traffic-light rule
4. 04Stripping the specifics
5. 05Three harmless details, one organisation extended reading
6. 06Shadow AI
7. 07A policy people actually follow
8. 08Try it: find out what actually applies to you try it
9. 09Quiz quiz

Section 18 was about your material going in. This is about what comes out: who owns it, whether you can use it, and when you have to say AI was involved.

1. 01Who owns what comes out overview
2. 02What the vendor’s terms actually give you
3. 03Using AI on things you did not write
4. 04The cases about training data
5. 05When it reproduces somebody else’s words
6. 06Checking what you publish
7. 07Saying AI was involved
8. 08Credit that means something
9. 09Quiz quiz

A straight map of AI regulation. The point is the shape: risk levels, and who carries which duty. When the specifics change, and they will, you can slot the new version into a structure you already understand.

1. 01Rules about uses, not about technology overview
2. 02The things that are simply banned
3. 03Levels of risk
4. 04Rules for the model makers
5. 05Which one are you?
6. 06Everywhere that is not the EU
7. 07Existing rules did not go away extended reading
8. 08A right to an explanation
9. 09Try it: map your own exposure try it
10. 10Quiz quiz

Hiring gets its own section because it is where the rules have landed hardest and where a bad tool discriminates against thousands of people at once, quietly, with a veneer of fairness.

1. 01Where it already sits
2. 02You are further in than you think
3. 03How these tools discriminate
4. 04Accurate on average, unlawful for some extended reading
5. 05Tools that claim to read people
6. 06The rules that already apply
7. 07What to ask a vendor
8. 08Applications written by AI
9. 09The human floor
10. 10Quiz quiz

How to judge an AI product on evidence rather than on a demonstration, and then how to choose: use the chat, buy or build, what “train it on our data” really means, and what a tool costs. Elective X3 turns the same discipline on things you build rather than things you buy.

1. 01Every demo is the best take overview
2. 02Twenty of your own cases
3. 03Building one properly
4. 04Questions a vendor should answer without flinching
5. 05Build, buy, or just use the chat
6. 06“Can we train it on our own data?”
7. 07What tuning a model is actually for
8. 08What it actually costs
9. 09Quiz quiz

How to get a group from “we did the training” to “the work is actually different”, and how to tell whether it worked without lying to yourself in the report.

1. 01Why training on its own changes nothing overview
2. 02Start with tasks, not tools
3. 03The four real objections
4. 04Answering each without brushing it off
5. 05The honest conversation about jobs
6. 06How it actually spreads
7. 07Shared prompts and stated norms
8. 08Measuring it without fooling yourself
9. 09Report what got worse extended reading
10. 10Quiz quiz

A look forward built around one documented episode rather than a list of predictions. The job of this section is to leave you with a way of reading AI news that keeps working after the specific story in it has aged.

1. 01Reading AI news without being sold to overview
2. 02Three trends with real evidence
3. 03A worked example: a capability arrives example
4. 04A worked example: the response example
5. 05A worked example: the sceptics example
6. 06What the episode teaches
7. 07When it leaves the screen
8. 08Two ideas that still apply
9. 09What is genuinely unresolved
10. 10What it means for your job
11. 11Try it: keeping up without drowning try it
12. 12Quiz quiz

Turning ten hours of reading into three things still happening next month. This course is not for knowledge. It is for a small number of changed habits, and this is where you write them down.

1. 01Why most training changes nothing overview
2. 02Try it: choose three tasks try it
3. 03Try it: write your own rules try it
4. 04Try it: set your workspace up try it
5. 05The checkpoints that make it stick
6. 06If you are rolling this out to other people
7. 07Quiz quiz

How to be the source an AI assistant names when somebody asks about your area. It is an elective because it is role-specific, but for anyone responsible for content it is the one with the clearest payoff.

1. 01The searches that never reach you overview
2. 02What it does to your numbers
3. 03What answer engines actually reward
4. 04Habits from the keyword era that now hurt
5. 05Writing that survives being cut up
6. 06The technical floor
7. 07Off your own site
8. 08Saying the same thing everywhere
9. 09Measuring it
10. 10Try it: run your question set try it
11. 11Quiz quiz

Building software and interfaces with AI. The oversight that keeps suggestions and generated code safe is reading them, and it does not survive the step to an agent that writes the change on its own. What an agentic coding tool is, the three guardrails that replace reading, the oversight dial, reviewing code you did not write, and where design work with these tools helps and where it misleads.

1. 01The ladder overview
2. 02Where the oversight breaks
3. 03What an agentic coding tool actually is
4. 04The harness decides what it can touch extended reading
5. 05Working with Claude Code, concretely
6. 06Guardrail one: what it may do without asking
7. 07Guardrail two: where it may do it
8. 08Guardrail three: checks that do not need you
9. 09Oversight is a dial, not a switch
10. 10Reviewing code you did not write
11. 11Designing interfaces in conversation
12. 12Making design work actually usable
13. 13Try it: set the guardrails for one repository try it
14. 14Quiz quiz

Section 22 applied the evidence discipline to products you buy. This applies it to things you build. A test set is the only thing that tells you whether last week’s change helped or quietly broke three cases.

1. 01Judging by eye stops working almost immediately overview
2. 02What one actually is
3. 03Build the set from real traffic
4. 04Unambiguous checks first
5. 05Graded checks, and using a model as the judge
6. 06Pin the versions, run on every change
7. 07Pick the measure that matches the harm
8. 08Reading failures beats reading the score
9. 09Watching it in production
10. 10Try it: build a twenty-case set try it
11. 11Quiz quiz

What it means to reach a model directly from your own program rather than through a chat window. Everything the chat window did quietly becomes your job, from keys and cost to memory and safety, and most of what breaks in production is some part of that job left undone.

1. 01What an API actually is overview
2. 02Keys, and why they matter
3. 03What a request actually looks like
4. 04The dials
5. 05The dials are already changing
6. 06Getting answers in a fixed shape
7. 07Letting it use your tools
8. 08Streaming
9. 09What it costs
10. 10Two ways to pay much less
11. 11What breaks in production
12. 12What you lose leaving the chat window extended reading
13. 13Choosing between them
14. 14Try it: make one call and read the answer try it
15. 15Quiz quiz