Vibe coding classes in Peru

Source: Stack Overflow, 2025 Developer Survey, AI section, read on 9 November 2026. The survey is global and answered by developers who chose to take part; it is not a Peru sample.

1. Ask for the code. The learner asks an AI tool to split a bill of a given number of cents between a number of friends. A very common style of answer divides, rounds, and gives everyone the same share. We wrote that version ourselves, below, to stand in for it.
2. State the rule before testing. Whatever the code does, two things must always be true: the shares must add back to the total, and no two people may pay more than one cent apart. Writing the rule down is the real skill; the code to check it is two lines.
3. Let the computer try. Generate 1,000 random bills (up to 50,000 cents) and groups (1 to 12 people), with a fixed random seed so the result can be repeated, and check the rule on every one.

import random

# Task given to an AI assistant (our wording): "split a bill of total_cents between
# `people` friends, in whole cents". This is the kind of answer it often gives:
def split_bill_generated(total_cents, people):
    share = round(total_cents / people)
    return [share] * people

# A version written to a rule we state first: the parts must add back to the total,
# and no two parts may differ by more than one cent.
def split_bill_checked(total_cents, people):
    base, extra = divmod(total_cents, people)
    return [base + 1] * extra + [base] * (people - extra)

def property_holds(parts, total_cents):
    return sum(parts) == total_cents and max(parts) - min(parts) <= 1

rng = random.Random(2026)
cases = [(rng.randint(1, 50000), rng.randint(1, 12)) for _ in range(1000)]

for name, f in [("generated", split_bill_generated), ("checked", split_bill_checked)]:
    failures = [(t, p) for t, p in cases if not property_holds(f(t, p), t)]
    print(name, "fails", len(failures), "of", len(cases), "| first:", failures[:1])

print("generated(1000, 3) ->", split_bill_generated(1000, 3), "sum", sum(split_bill_generated(1000, 3)))
print("checked(1000, 3)   ->", split_bill_checked(1000, 3), "sum", sum(split_bill_checked(1000, 3)))

generated fails 762 of 1000 | first: [(7807, 6)]
checked fails 0 of 1000 | first: []
generated(1000, 3) -> [333, 333, 333] sum 999
checked(1000, 3)   -> [334, 333, 333] sum 1000

1. Ages 8 to 12, no code. Share 10 sweets between 3 friends so that nobody gets more than one extra. How many ways are there? What is the rule?
2. Ages 13 to 17. Ask an AI tool your age allows for a function that finds the middle value of a list. Write the rule it must obey, then test it on 1,000 random lists, including empty ones.
3. Adults. Pick a function from a project you vibe coded. Write one property that must always hold, and test it on random inputs. Report what you found in three English sentences.
4. Everyone. Explain why testing with "nice" numbers is not enough, using this lesson as your example.

Competitions make the same point from the other side. Peru's Olympiad in Informatics, OPI, which UTEC runs free for all secundaria students in the country, prohibits generative AI for solving problems in its 2025 rules. A teenager who wants to compete has to reason and code alone. We are not connected with UTEC or the OPI; we simply make sure our vibe coding students can still code without the vibe.

If you would like to read about vibe coding in Spanish first, our vibe coding curso en inglés page explains it for Peruvian families with a different lesson. Learners who want to go deeper into one language can continue with Python in English or web development in English.