Start Here: What You Can Do with AI and the Work Loop

Two colleagues get the same task on the same Monday: summarise a 30-page report for a meeting. One pastes the whole thing into an AI assistant, types "summarise this", and forwards the answer. The other tells the assistant who the summary is for, asks it to quote the passages it relied on, then checks three numbers against the original before sending.

Both used AI. Only one of them can be sure the summary is right. The difference is not which tool they chose. It is a set of habits, and those habits are what this course teaches.

• What you can realistically do with AI today, and what you cannot.
• How this course is organised and how to pick a path through it.
• The Work Loop, a six-step habit you will apply in every module.
• A simple way to decide how much checking an AI answer needs.
• How to set up a free, safe starting environment in about fifteen minutes.

Modern AI assistants are built on large language models. They are very good at working with language: drafting, rewriting, explaining, summarising, translating, brainstorming, and writing and explaining code. They are weaker, or unreliable, at anything where being exactly right matters and they cannot check: precise numbers, recent events, obscure facts, long chains of calculation, and claims about specific people or documents they have not been shown.

That mix explains most of the good and bad stories you hear. In practice, a careful user can expect help like this:

Nothing in this table is a promise. Results depend on the tool, the task, your instructions and your checking. This course will not claim that AI will make you a better developer, a faster worker or a more employable person. It teaches the practices that give you a fair chance of those outcomes and protect you when things go wrong.

This is not a list of the best AI tools. Tool lists go stale within months, and they teach you where the buttons are, not how to think.

Instead, the course teaches transferable skills: understanding what the technology is, asking clearly, supplying context, verifying output, improving through feedback, and automating only what deserves it. Where a lesson uses a current product as an example, that is flagged as an example. When a feature is likely to change, the lesson is built around the underlying workflow, so it still works after the interface moves.

It is also not a technical course on building AI models. You will learn enough about how models work to make better decisions, and no maths is required.

Here is the habit that ties the whole course together. Whenever you use AI for a task that matters, walk through six steps.

You may notice this resembles the sequence understand, ask, give context, work together, verify, improve, automate. It is the same idea. "Understand" is Module 1, which sits underneath every step. "Decide" is added at the front because the most expensive mistake is using AI on a task it should not touch, or trusting it without a plan to check.

Say you must give a ten-minute talk to your class on how UPI payments work. Let us walk the loop once.

1. Decide. AI is a good fit for structuring a talk and explaining ideas simply. It is a poor judge of exact facts, such as transaction limits or dates, which change and may be misremembered. So you will use it for structure and wording, and check every fact yourself.

2. Describe. A weak first request:

Make a presentation on UPI.

A better one states the goal, audience, constraints and format:

I am a second-year engineering student giving a 10-minute talk to classmateswho use UPI daily but do not know how it works. Suggest an outline of 5 to 6sections. For each section give one sentence on what to say and one questionI could ask the audience. Do not include any figures or dates; I will addverified ones myself.

3. Ground. You paste in your own rough notes, and a paragraph from the official source you plan to rely on, and say: "Use only these notes for factual statements. Put everything else under 'to verify'."

4. Check. An example of the kind of response you might get (illustrative only, yours will differ):

Example response (illustrative)1. Why UPI exists - ask: "How did you pay for things before it?"2. Who is involved: you, your bank app, the merchant, the banks......To verify: current per-transaction limits, which organisation runs the network

The "to verify" list is the useful part. You look each item up in an official source and write down what you found. If the assistant states a figure anyway, you treat it as a claim to check, not a fact.

5. Refine. You notice section 3 is too technical for your audience. You say: "Rewrite section 3 for someone who has never heard of a bank server, using an analogy involving a postal system, and keep it to three sentences." One targeted instruction beats starting over.

6. Systemise. You will give more talks. You save the improved request as a template with blanks for topic, audience and length. Next time it takes two minutes. That small template is the seed of the reusable workflows in Modules 2 and 9.

Notice what happened. The assistant did the part it is good at (structure and wording). You did the part it is not reliable at (facts), and you used the loop to keep those two jobs separate.

You cannot verify everything, and you should not try. Match the effort to what happens if the answer is wrong.

You will use this ladder again in How to Fact-Check AI-Generated Answers, where it grows into a full verification routine.

You do not need a subscription. Here is a sensible starting kit.

1. Pick one general-purpose AI assistant with a free tier. Any mainstream one is fine for this course. Using two, once you are comfortable, helps you notice when answers differ.
2. Open its privacy or data settings and read what it does with your conversations, including whether they can be used to improve its models. Settings and defaults differ by product and change over time, so look at the current page rather than trusting a summary, including this one. Privacy and Security: What Not to Paste, Connected Tools and Prompt Injection goes deeper.
3. Create a notes file called something like ai-playbook.md. Every time something works, paste the request and a one-line note on why. This becomes your personal library in Module 9.
4. Set one rule now: do not paste passwords, API keys, personal identifiers, customer data or confidential work documents into a consumer assistant. If you are unsure, do not paste it.

Reading in order is best, because later articles build on earlier ones. If you are short on time, follow a track after Module 2:

• Student: Module 3, then articles 26 and 28, then Module 9.
• Working professional: Module 4, then 26, 28 and 30, then Module 9.
• Developer: Module 5, then 26, 28 and 30 to 32, then Module 9.
• Job seeker: Module 6, then 26 and 28, then Module 9.

Modules 1, 2 and 7 are core for everyone.

1. Read for the idea first, without a tool open.
2. Try the examples with your own tasks, not mine. The point is transfer.
3. Attempt every exercise before opening its solution.
4. Add one reusable item to your playbook per article.
5. Revisit your playbook weekly and delete what you never use.

• Judging AI by one bad prompt. A vague request gets a vague answer. That says little about the tool.
• Trusting fluent writing. Smooth prose is not evidence of accuracy. Module 1 explains why.
• Collecting tools. Three assistants you barely know are worth less than one you use well.
• Skipping the check step because the answer looked right. The cases where it looks right and is wrong are the dangerous ones.
• Pasting sensitive material to save time. Convenience is the usual way data leaks.
• Expecting the course to guarantee results. It cannot. It can improve your odds and reduce your risks.

I am a final-year B.Tech student applying for junior backend developer roles.Below is my current resume text. Suggest improvements to clarity andstructure only. Constraints: do not add any skill, project, number orachievement that is not already in the text; if a bullet is vague, ask mea question instead of guessing. Format: show each original bullet, then yoursuggested rewrite, then one sentence on why.[paste resume text]

• Next: AI, Machine Learning, Generative AI and LLMs: The Map clears up the vocabulary so the rest of the course makes sense.
• For a shorter, lighter introduction to how chatbots work, see How Do AI Chatbots Like ChatGPT and Claude Actually Work?.
• If you want prompting first, you can jump to Prompt Fundamentals: Goal, Context, Constraints and Format, then come back to Module 1.