<!-- Markdown version of https://pococo.ai/learning/ for LLMs and AI agents. -->
# Learning · pococo

Three-minute AI lessons. One skill per lesson, to try on your own work as you read.

## What a lesson is

One skill start to finish: what to open, what to type, what you should see back, and what to do when it comes back wrong. Written to follow along with your own work open beside it. Three minutes is the target and five the ceiling; anything longer is split in two. No theory, no tool comparisons, no reading lists.

## Lesson 01

"I get a wall of text I can't use," in the four steps every lesson follows:

1. **Open** the AI you already use, and the question that got you five paragraphs.
2. **Type:** ask again, and say what shape you want back: "Answer in three bullet points, one line each. No intro, no summary."
3. **You should see** three short lines you can act on.
4. **If it comes back wrong,** paste a short sample of the layout you want (made-up words are fine) and ask it to match.

## Lesson 02: I don't know what to use it for

Choose one recurring task with a result you can check.

1. **Open:** A blank note. List three small tasks you repeat, using descriptions you can share. For example: turn meeting notes into an action list, shorten an update, or organize a to-do list.
2. **Type:** Ask your AI to help you choose a first experiment. Here are three tasks I repeat: [your list]. Suggest one small task where you can help draft or organize something. Tell me what to provide and how I can check the result. Leave the final decision to me.
3. **You should see:** One concrete task, the input it needs, and a simple check. Try it with a made-up example or material you are allowed to share. Compare the result with what you expected.
4. **If it comes back wrong:** Make the task smaller: “Use only these notes to draft a five-item checklist.” Skip tasks whose result you cannot check. Keep the part that helps and do the rest yourself.

## Lesson 03: It sounds confident and I can't tell if it's right

Check one important claim against the original source.

1. **Open:** An AI answer and the public or approved source it is meant to describe. Choose one checkable claim that matters, such as a date, amount, or requirement.
2. **Type:** Ask it to separate the claim from the evidence. For this claim: [claim], identify the exact supporting passage in the source I supplied. Separate what the source says from your inference. If the source does not support it, say so.
3. **You should see:** A passage you can find in the source, or a clear statement that support is missing. Open the original yourself and compare the words, date, and context. Check calculations independently. A plausible-looking citation is not enough.
4. **If it comes back wrong:** Mark the claim unverified and leave it out of your work until you can check it. Ask for a version limited to supported facts. Checking one claim does not verify the whole answer.

## Lesson 04: It doesn't sound like me

Use a short sample of your own writing to show the tone.

1. **Open:** A draft you want to improve and two or three sentences you wrote yourself. Use a sample you are allowed to share, with private details removed.
2. **Type:** Give the sample and a narrow editing instruction. Here is a sample of my writing: [sample]. Edit this draft to match its sentence length, tone, and level of formality: [draft]. Keep the facts and intent. Do not add claims or promises. Show me the revised version.
3. **You should see:** The same message in a voice closer to yours. Read it aloud and check names, dates, commitments, and anything the recipient needs to do. Keep only edits you would actually make.
4. **If it comes back wrong:** Name one mismatch: “Use shorter sentences,” “This is too formal,” or “Keep my opening line.” Revise that part. A style sample helps guide the draft; you still choose the final words.

## Lesson 05: I don't know if my work is safe to put in

Decide what to share before you paste anything.

1. **Open:** The task you want help with in your own notes. Keep the source material there for now. Identify the minimum information needed, and check which AI account your workplace permits for that kind of data.
2. **Do:** Make a practice version using invented names, numbers, and text. Check your workplace rules and the data-use information for the exact product and account you use. Account settings and agreements can differ. Ask the responsible person if the allowed use is unclear.
3. **You should see:** A small practice input with no private source material, and a clear decision about what you are permitted to share. You can use the practice input to test the workflow while keeping real data out.
4. **If it is still unclear:** Do not paste the real material. Use the invented example or ask the AI for a blank template. Removing names alone may not make a document anonymous, and a training setting alone does not establish permission to share it.

## Requests

Email hello@pococo.ai with a problem or a lesson request. A question may become a lesson, and the most-requested get written first.

Bigger than a lesson: https://pococo.ai/consulting/ covers it, including team training on your own work.


## Worked examples

### 01: From notes to next steps

Made-up notes: Mina will send the draft on Tuesday. Theo will review it on Thursday.

Ask for a table with three columns: action, owner, due date. You can now check each row against the notes.

Try it on your work: Keep the same notes and ask for a two-line email instead. Notice how changing the format changes what you can do with the answer.

### 02: Start with a weekly update

You repeat three tasks: summarising progress, naming files, and planning a budget.

Try summarising three made-up progress notes first. Check that every item came from the notes and that nothing was added.

Try it on your work: Pick a task you can finish and review in one sitting. Write down what a useful result would look like before you ask.

### 03: Check the date, not the confidence

A made-up event listing says “12 October”. The AI summary says “21 October”.

The original listing supports 12 October. Correct the date and check the other details separately.

Try it on your work: Choose one name, date, or number in an answer. Find it in the original source before you keep it.

### 04: Make the message sound like you

Your sample: “Thanks for this. I’ll take a look tomorrow.” The draft: “I am delighted to confirm receipt of your correspondence.”

A closer version: “Thanks for sending this. I’ll take a look tomorrow.” Check that tomorrow is a commitment you intend to make.

Try it on your work: Give one specific direction, such as “keep my opening line”. Read the revised draft aloud before you use it.

### 05: Keep the structure, invent the details

You want to turn a customer request into a checklist. The real message includes a name, address, and order details.

Practise with: “Customer A ordered two blue mugs. One arrived damaged. Draft a checklist for reviewing the request.”

Try it on your work: Write a tiny invented example that preserves the task. Keep the original message in your own notes while you try the workflow.
