/01

AI 101

What AI is, what it is good at, where it fails, and the five words you actually need to know.

Start here

8 min

Open +

You do not need to become technical. You need a simple mental model so you know when AI can help and when it is bluffing.

Leave knowing / Understand AI well enough to use it

The one-minute version

Modern AI is very good at finding patterns in information and producing a useful next output. That output might be words, code, an image, a classification, a summary, a plan, or a choice between options.

It can feel like a person because language is the interface. Do not confuse a good interface with guaranteed truth, judgment, permission, or understanding.

AI is useful by default. It is not truthful, current, private, or authorized by default.

The simple model

What happens when you ask AI for something.

You give it a job.

A question, document, image, spreadsheet, codebase, or instruction becomes the input.

Output / Input

The model makes a best next move.

It uses patterns learned during training plus the context you gave it. It can still be confidently wrong.

Output / Model

Tools can give it hands.

With permission, an AI system can search, read files, query data, write code, update software, or call another service.

Output / Tools

Checks decide whether to trust it.

A person, rule, test, source, permission check, or other verifier should confirm important outputs before they matter.

Output / Proof

Five words

The only vocabulary you need to start.

Model

The pattern engine. It turns the context it can see into a likely useful response.

Prompt

The job you give the model. A good prompt is really just a clear brief.

Context

The information the AI can see right now: instructions, files, history, data, and tool results.

Tool

A real capability the AI can use, like search, a calendar, a database, code, or email.

Agent

AI plus tools, rules, memory, and a loop that lets it work through more than one step.

Good first jobs

Start where language and repetition already eat time.

Turn one thing into another.

Meeting notes into actions. A long report into a brief. A rough idea into a first draft.

Find patterns.

Group customer feedback, compare documents, flag repeated issues, or explain what changed.

Make a first pass.

Draft emails, plans, code, job descriptions, FAQs, briefs, formulas, or test cases.

Explain hard things simply.

Ask it to teach you a document, process, chart, policy, or technical idea at the level you need.

Keep a human

Do not hand these over just because the demo looked impressive.

Final decisions that can seriously affect a person, customer, employee, or business.

Facts that matter unless the answer is tied to a source you can inspect.

Passwords, private keys, regulated data, or confidential information in an unapproved AI tool.

Irreversible actions unless permissions, limits, logs, and a recovery path are already designed.

Anything where nobody can explain who is accountable if the AI is wrong.

Try it today

One useful first exercise.

Use a real piece of work you already have. Do not invent a fake AI project.

Try the template

I am going to paste notes from a real meeting.

Turn them into:
1. Decisions that were made
2. Actions, with an owner if one is named
3. Open questions
4. Risks or blockers
5. A five-sentence summary I can send to the team

Do not invent missing facts. Mark anything uncertain as "Needs confirmation."

Here are the notes:
[PASTE NOTES]

Notes / Subscribe

No schedule. No content calendar. A note goes out only when there is something worth understanding.