Scenario quizzes for beginner AI engineers
Debug a broken prompt, pick the right tool call, iterate on a vibe-coding step, one short quiz at a time
Step 1 of 15 · Prompting
Debug a broken prompt
Which edit to the prompt most reliably gets the agents their three bullets?
"Be concise and keep it short so agents can skim it." to the end of the system prompt."3 bullets under 15 words: the problem, what they tried, what they want."How it works
Three moves in every step
- 01
Read the scenario
A short real-world scenario, not a theory question: a broken prompt, tool call or AI-built change, with the part at fault highlighted
- 02
Pick the fix
Choose the edit that solves it from three or four answers
- 03
See why it works
Every answer explains why it is right or wrong, then the next question follows
The path
One path in three parts
Prompting first, then tool calling, then vibe coding. Finish a step to open the next one.
15steps in all
See the whole pathPrompting
5 steps
Tool calling
5 steps
Vibe coding
5 steps
Example
3 questions in this step
- 1Which edit to the prompt most reliably gets the agents their three bullets?
- 2Why did the original prompt produce a paragraph in the first place?
- 3You have rewritten the prompt. How do you check that the fix actually works?
Example
Debug a broken prompt
Your support team gets hundreds of customer emails a day, so you built a summariser that runs on each one before an agent opens it. Agents complain the summaries are as slow to read as the emails: the model writes a long paragraph, repeats the customer's apology and buries what they actually want. The agents asked for three short bullets: the problem, what the customer already tried, and what they want us to do.
Prompt
System: You are a helpful assistant. Summarize the following customer email so our support agents understand it. User: {email_text}
Model output
The customer, who has been a loyal user for several years and expresses some frustration, is writing to let the team know that they have been experiencing difficulties with their recent order, which arrived damaged. They mention that they have already tried contacting support through the chat, and they would appreciate it if the team could look into the matter and possibly arrange a replacement or a refund, whichever is easier, as soon as possible.
Which edit to the prompt most reliably gets the agents their three bullets?
- Add
"Be concise and keep it short so agents can skim it."to the end of the system prompt. - Spell out the format:
"3 bullets under 15 words: the problem, what they tried, what they want." - Switch to a bigger, more capable model, since this one clearly isn't following the instructions it was given in the prompt.
- Set
temperatureto 0 so the model stops improvising and sticks closely to the instructions it was given.
Right answer: choice 2
The model can only follow a format you state. Giving the count, a length limit and what each bullet covers turns a vague wish into a checkable instruction.
Step 1 is a support‑email summariser that rambles. Go fix it
- Free
- No sign-up
- Runs in your browser, nothing to install
- Your progress stays in this browser
Opens your next question straight away