Learn to use AI by making requests specific and checking the results yourself.
Choose a question that interests you and create a real output.
If you want to understand the principles first: What Is Generative AI? Understand How Answers Are Generated and Why Hallucinations Happen
Build a request that clearly includes your goal and constraints.
Practice with hints and review of wrong answers.
Create a weekly plan that respects your time limits.
Learn how to compare a summary against the original source.
Start vibe coding with Claude Code by writing the requirements first.
Compare repetitive tasks and pick your team's first use case.
Step by step, with examples
This hands-on exercise covers reviewing long material by dividing it into multiple chunks, while checking the document ID and version, overlap at chunk boundaries, different deadlines, and the full scope. Rather than focusing on splitting itself, it emphasizes returning to the source text to verify the merged result.
This hands-on exercise compares two AI answers using the same input and the same prompt, checking them against predefined factual criteria rather than writing preference or brand impression. Separately from the score, unsupported additional claims are checked as a gate for whether the result is suitable for publication.
This exercise uses AI to compare V1 and V2 of the same event notice and classify five comparison items as changed, added, deleted, or unchanged. You will learn how to record only differences that actually appear in the documents without assuming why a deleted sentence disappeared.
This exercise turns a short notice into six FAQ entries. Questions supported by the source receive source IDs, while topics such as fees, refunds, and waitlists that do not appear in the notice are left as needs confirmation instead of being answered from assumptions.
This exercise classifies six short free-text comments into predefined themes and allows multiple labels when one response contains more than one topic. It also explains why the total number of labels can exceed the number of respondents and how to avoid unsupported conclusions about satisfaction or emotional intensity.
This exercise rewrites four fictional borrowing rules in clearer language while preserving the registered-member requirement, the 7-day deadline, the prohibition on extending reserved materials, and the rule that non-reserved materials may be submitted for one extension request without guaranteeing approval.
This exercise shows how to break a vague research goal into factual questions, experience questions, and decision questions, then plan what to verify and where to look. Instead of asking AI to invent search results, you first organize the sources to check and the items that are still unknown.
This exercise shows how to design a 15-minute user interview for improving a fictional application form. You will revise leading questions into neutral questions about recent real behavior and learn not to treat predicted responses as evidence before any interviews have been conducted.
This exercise shows how to define constraints such as participant count, time, printed materials, purchasing limits, and available equipment before asking AI for 4 distinct alternatives. Instead of deciding which idea is best, you check whether each option satisfies the constraints and learn how to revise the request when the constraints change.
This exercise shows how to prepare 6 expected audience questions with source evidence, short answers, and a verification plan when the answer is unknown. It does not cover creating a presentation outline or AI voice evaluation; the focus is on answering questions without overstating limited evidence.
Learn how to turn short source text into an AI PPT presentation structure without adding unsupported facts, create a prompt for a 6-slide outline with one key message, evidence, speaker notes, and visual suggestions, and manually review numbers, interpretations, images, and sensitive information.
Learn what generative AI does and how text-generating AI produces responses by continuing tokens from the context it has been given. Because fluent language and factual accuracy are not the same thing, you will also practice separating confirmed information from unknown information and checking the result yourself.
When introducing team collaboration AI for the first time, it is easier to manage if you compare tasks where multiple people repeatedly collect and verify information rather than starting with eye-catching features. Using 6 fictional tasks from Team Haneul, this guide describes expected benefits qualitatively, checks review burden and sensitive-information risk, and then narrows down the first use case. AI organizes the candidates, but the team makes the final choice.
When creating AI meeting notes, instead of polishing only one person's notes, practice combining records from multiple attendees and organizing decisions, action items, TBD items, and differences between records together with source tags. Using fictional Team Haneul meeting notes from 2026-10-20, this guide shows how to create a draft for team sharing without forcing different wording into a single version.
Practice using AI to collect weekly updates in different formats from each team member and reorganize them into completed work, work in progress, delays, dependencies, help needed, and items needing confirmation. Using Team Haneul reports from the same period, this guide creates an AI weekly report example that preserves both reference times and sources when numbers differ instead of deleting one value.
This guide explains how to reorganize scattered notes into background, current status, material locations, next actions, cautions, and TBD items so someone else can continue the work immediately when the owner is absent. Using a fictional case in which Seoyeon's promotion and registration work is handed over to Junho, it creates both an AI work handover document and questions the recipient should verify.
Practice comparing confirmed facts, evidence, opposing views, and unverified assumptions in one framework without letting AI make the conclusion when team members disagree. Using Team Haneul’s Venue Option A and Option B for the workshop, we identify the decision owner and deadline clearly and leave items for the team to verify before making the final judgment.
Practice turning a short work request from a team lead into a draft request the team can agree on again, separating confirmed requirements from missing conditions instead of letting AI fill in the gaps arbitrarily. Using Team Haneul’s request, 'Please prepare some workshop materials,' we leave completion criteria, included and excluded scope, and submission conditions as clarifying questions.
Learn how to turn team collaboration AI into a team asset that is reviewed and revised together rather than a set of individual prompts. We follow Team Haneul as it improves an AI meeting-notes prompt from v1.0 to v1.1 and checks it again with the same test input.
Learn how to check facts, omissions, sensitive information, owners, and approval status by role before sharing an AI weekly-report draft with the team. Separate the author, reviewer, and approver and record what evidence was checked so AI output is not mistaken for the team’s official conclusion.
Instead of treating vibe coding as simply 'tell AI what you want and it builds everything for you,' this article frames it as a workflow in which a person defines the requirements and verification criteria, then checks the result produced by AI. Using a fictional household-expense CSV summation tool, it organizes the input, output, things not to do, and verification method into a one-page note.
Check the Claude Code installation command and account requirements for your operating system, then start a first session in a fictional practice folder. Instead of generating code immediately, this article focuses on asking only three questions in plan mode to inspect the folder and CSV, then exiting safely.
Use the fictional expenses.csv and requirements note from the previous articles to practice building a first small Python tool. Put the input example, expected output, and constraints in the request, then run the completed summarize.py separately in the Python tool in GPT chat—not in Claude Code—to verify the result.
Put the run command, data rules, prohibitions, and verification method that were being repeated in every request into the project's CLAUDE.md in a concise form. Distinguish the `/init` function that creates a draft from the purposes of different file locations, and complete a 35-line rules file for the expense CSV example.
Using the example of adding a `--by-category` option to an already working monthly expense summary tool, learn how to review a change plan first in Claude Code's plan mode. Fix the scope of edits and output format before implementation, then run the completed code after approval to verify both the existing monthly totals and the category totals.
Create a git baseline before Claude Code edits files, then read the actual changes with `git status` and `git diff`. You can ask AI to explain the diff, but do not rely on the explanation alone: review the diff and execution results yourself, then stage and commit the changes directly.
Instead of telling AI only that 'there is an error,' reproduce the problem with the same input, narrow down the cause from the actual exception message, and request only the minimum necessary fix. Using a CSV where an amount contains a comma, reproduce a ValueError in Python and rerun the corrected code to verify the fix.
Assuming Claude Code can read and modify files and run commands, organize the safety checks beginners should make before, during, and after work. Connect the differences between permission modes, changes checkpoints cannot restore, the separate role of git, and the rule of not exposing secrets into one practical checklist.
Learn how to get easier-to-check results by giving AI the goal, confirmed facts, prohibited assumptions, and desired format instead of simply asking it to write something. Using a neighborhood book club notice as an example, you will check whether AI invented missing information and then use a follow-up request to adjust only the tone while keeping the facts unchanged.
Learn to use AI for practice by having it present one problem at a time, wait for your answer, and then provide hints and explanations instead of immediately giving the answer. Using three percentage problems, you will check which value belongs in the denominator and independently verify AI explanations with your own calculations.
Learn to calculate the time you actually have available and the time required for your tasks before asking AI to place work into a schedule. The exercise also covers a situation where Wednesday suddenly becomes unavailable and shows how to verify that AI does not invent extra time that does not exist.
Turn “automate this” into an executable task description. Attach a synthetic sample with no sensitive data and a hand-checked expected result to complete a request for a script that totals work logs by team.
Check an aggregation function with four rows you can calculate by hand and 12 unit tests. Verify not only normal values but also empty input, zero, decimals, and invalid input.
Check an AI report against its source data, recalculate totals and rates, and use a copyable review prompt to flag unsupported claims and overstated causes before submission.
Treat an AI-generated regex as a draft, not a finished rule. Build a small synthetic test table, compare expected and actual matches, revise the pattern, and save a review report before using it on real data.
Treat an AI-generated SQL query as a draft and test it on a tiny SQLite database with known answers. Compare the AI result with a manually calculated expected table before using the query on real data.
Show an AI a few examples of the exact output structure you want, then validate the returned JSON before using it. A small synthetic ticket dataset demonstrates how formatting examples and automatic checks work together.
Break an AI summary into individual claims, link each claim to specific source sentences, and mark unsupported or overstated statements before reuse. A small synthetic document shows why fluent summaries still need evidence checks.
Review an AI-drafted work email against the original facts before you send it. This example checks names, quantities, prices, dates, recipients, tone, and wording that could accidentally create a promise.
Use an AI chat assistant to turn a meeting transcript into structured minutes, then verify every decision, owner, deadline, and action item against the transcript. A small synthetic example and Python script show how to make that review repeatable.
Do not trust an AI-suggested Excel formula after testing only normal rows. Use a small synthetic sheet to test blanks, zero, text, and dates, then cross-check the intended results with a simple Python script.
Review an AI-translated work document systematically instead of checking only whether it sounds natural. Use a small synthetic example to compare numbers, units, names, negations, and selected back-translations against the source.
Use an AI chat assistant to propose explicit data-cleaning rules, apply only approved rules to a copy of a small CSV, and create a cell-level before/after log. The example is synthetic so you can inspect every transformation yourself.
An AI can describe a table fluently while still getting totals, percentages, averages, or changes wrong. This guide uses a tiny synthetic sales table and a Python script to recompute each numerical claim before you reuse it.
Recurring AI tasks become easier to review when prompts follow a consistent structure. This guide shows how to build a small prompt template library with variables, examples, and a checklist for each task.
AI can turn unstructured notes into structured JSON, but the result still needs validation. This guide uses a tiny synthetic example and a Python schema check to verify required fields, types, allowed values, and basic consistency.
AI can label support tickets quickly, but the labels should be measured against a small hand-labeled reference set before being trusted. This guide uses a tiny synthetic example, compares AI predictions with known labels, and calculates accuracy plus a confusion matrix in Python.
Remove or replace obvious personal data before sending text to an AI chat assistant, then review both the masked text and a detection report for anything the script missed. This tutorial uses a small synthetic example and keeps the original file unchanged.
AI-generated code can look complete while still making assumptions about inputs, edge cases, dependencies, and failures. This guide uses a tiny synthetic CSV example to review the code systematically before relying on it.
AI answers about documents are easier to verify when every claim includes line-number citations. This guide uses a tiny synthetic note, asks an AI chat assistant for cited answers, and checks each citation range automatically with Python.
AI can turn several paper abstracts into a compact comparison table, but each cell should be traceable to the source text. This guide uses three tiny synthetic abstracts, asks an AI chat assistant to build a table, and checks every table cell against the abstracts before accepting it.
AI-generated references can look plausible even when a link is missing or the cited page does not support the claim. This guide uses a small synthetic example to check both URL existence and whether the referenced page contains evidence related to the AI's claim.
Use AI to draft a weekly report from CSV work logs and Python to reconcile hours and completed tasks. Copy a prompt that separates completed, ongoing, planned, and blocked work.
Try a short search term such as CSV or AI.
All examples are for practice. Automotive data is synthetic and used for illustration.