Canada · Vibe coding and AI agents · Teens and adults

Vibe coding and AI agents classes in Canada

Describing what you want and letting an AI write the code has a name now, and Collins Dictionary chose it as its Word of the Year for 2025. AI agents go a step further: they decide which tools to call and in what order. Both are useful, both fail in predictable ways, and both have age rules that most courses never mention. This page explains the terms from their sources, sets out which tools a Canadian teenager may use and which are for adults only, lists what agents can and cannot do, and builds a small, honest agent on City of Ottawa park data whose every answer is checked.

Teaching live online since 2020 · 10,000+ students · separate courses for teens and adults, with every age rule respected

In short

Vibe coding and AI agents classes in Canada teach teenagers and adults to build software with AI help, then read, test and correct what the AI produced, and to design agents whose answers can be checked. Teens and adults learn separately, within each tool's age rules. Modern Age Coders teaches them live online, in a batch or one to one.

Separate courses by age

Three courses: one for teens, two for adults

Teenagers and adults may legally use different tools, so they learn in different courses.

What the words mean

Vibe coding, agents and MCP, defined by their sources

These terms are used loosely in advertising. Here they are as the dictionary, the agent builders and the protocol's own site define them, read on 22 October 2026.

Definitions from primary sources (read 22 October 2026)
TermSourceWhat the source says
Vibe codingCollins Dictionary, Word of the Year 2025 (November 6, 2025)"Coined by AI pioneer Andrej Karpathy, vibe coding refers to the use of artificial intelligence prompted by natural language to write computer code."
WorkflowAnthropic, Building effective agents (Dec 19, 2024)"Workflows are systems where LLMs and tools are orchestrated through predefined code paths."
AgentThe same articleAgents "are systems where LLMs dynamically direct their own processes and tool usage".
Agentic coding toolClaude Code documentation"Claude Code is an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools."
MCPmodelcontextprotocol.io"MCP (Model Context Protocol) is an open-source standard for connecting AI applications to external systems."

The difference that matters in practice

In a workflow, a programmer decides the steps and the model fills in parts of them. In an agent, the model decides the steps. That one change explains most of what goes right and wrong: an agent can solve a problem nobody scripted, and it can also wander, repeat itself, or act on a mistaken guess.

Vibe coding sits at the human end of the same scale. A person describes the result, the AI writes code, and the person decides whether to keep it. The skill that separates a good vibe coder from a bad one is reading what came back.

What we teach first

Before any agent, learners write the tools themselves: small functions with clear inputs, outputs and failure messages. An agent is only as reliable as the tools it calls and the checks wrapped around its answer.

Anthropic's own advice points the same way: "we recommend finding the simplest solution possible, and only increasing complexity when needed." Many problems never need an agent at all.

Sources: Collins, Word of the Year 2025; Anthropic, Building effective agents; Claude Code overview; Model Context Protocol. All read 22 October 2026. We are not connected with any of these organisations.

Who may use what

Which AI coding tools a teen may use, and which are adults only

This is where many vibe coding courses go wrong: they hand every student the same tools. The companies' terms do not allow that. Here is what each says, and how our two age groups follow it.

Age rules for AI coding tools, from each company's terms (read 22 October 2026)
ToolWhat the terms sayTeens 13 to 17Adults 18+
Claude, including Claude Code (Anthropic)"You must be at least 18 years old or the minimum age required to consent to use the Services in your location, whichever is higher." (effective Oct 8, 2025)Not allowed; never recommendedAllowed
ChatGPT and other OpenAI services"You must be at least 13 years old"; "If you are under 18 you must have your parent or legal guardian's permission to use the Services." (effective January 1, 2026)Only with a parent's or guardian's permissionAllowed
GitHub and GitHub CopilotGitHub: "A User must be at least 13 years of age." Copilot Student is "Available for free to verified students, teachers, and open-source maintainers".Allowed from 13; free if verified as a studentAllowed
The Canadian Computing CompetitionThe CCC rules forbid generative AI and say "AI-based tools within an IDE cannot be used."Practise without assistants before 17 February 2027Not applicable

How our teen classes work

Teenagers build with tools their age allows, chiefly GitHub Copilot for verified students and, where a parent has given permission, OpenAI's tools. No teen class uses Claude or Claude Code, and no page of ours suggests them to anyone under 18.

For a teen, the point of vibe coding is not speed. It is learning to judge code: why the suggested function works, where it breaks, and what the AI did not consider.

How our adult classes work

Adults work with agentic tools such as Claude Code and OpenAI's Codex, on real repositories, with version control so that every change an agent makes can be inspected and undone.

The adult course ends with a phase on building, securing and shipping, because reviewing and testing agent output is where the real work is.

Sources: Anthropic Consumer Terms; OpenAI Terms of Use; GitHub Terms of Service; GitHub Education; CCC rules. All read 22 October 2026; terms change, so check before signing up.

Honest limits

What AI agents can do, and what they cannot

Agents are sold as autonomous workers. The people who build them are more careful. This table combines Anthropic's published guidance with what our own small agent below showed.

Agents: strengths and limits (Anthropic's guidance, quoted; our test results, labelled)
QuestionWhat the evidence says
Can an agent handle a task nobody scripted?Yes. Anthropic: agents suit cases where "flexibility and model-driven decision-making are needed at scale".
Is an agent always the right choice?No. Anthropic recommends "finding the simplest solution possible, and only increasing complexity when needed."
What does autonomy cost?Anthropic: "higher costs, and the potential for compounding errors."
How should agents be tested?Anthropic: "We recommend extensive testing in sandboxed environments, along with the appropriate guardrails."
Can an agent state something its tools never returned?Yes. Our stand-in plan below names a park, "Maple Leaf Dog Run", that is not in the city's file. Only the check catches it.
Does an agent understand data it was not told about?No. The Ottawa file codes dog rules as 0 to 4. Nothing in a code says that 0 means off leash and 4 means not designated; the agent knows only what its tool description says.

Good at

Repetitive changes across many files, first drafts of tests, exploring an unfamiliar codebase, and joining tools a person has already made reliable.

Weak at

Knowing when it is wrong, respecting a rule nobody wrote down, and stopping when the task is unclear. It will often produce a confident answer instead.

Needs from you

Clear tools, a test you can run, version control to undo its changes, and a check that compares its answer with what its tools actually returned.

A real agent example

An Ottawa parks agent whose answers are checked

The City of Ottawa publishes its parks and greenspace under a licence that allows commercial reuse, including a coded rule for dogs in each park. We built a small agent loop around it to show every part an agent needs: tools, a loop that runs them, and a check on the final answer. The code and every number are ours, run on the city's file downloaded on 22 October 2026.

The question an agent is asked: "Which parks in Kanata North let my dog off leash?" The file has 1,378 rows across 24 wards, with columns including NAME, WARD_NAME, DOG_DESIGNATION and DOG_DESIGNATION_DETAILS, the city's own wording for each park's rule.

What the DOG_DESIGNATION codes mean, read from the city's own detail text (our tally of the file)
CodeThe city's wording, most common formParks
0"Dogs may be off leash." (with a line about the handler's control)175
1"Dogs are allowed but must be on leash at all times."356
2Area-specific, for example off leash in a fenced area and on leash elsewhere69
3"Dogs are not allowed in the park."119
4"The park has not been designated." (then on-leash wording)631
BlankNo code given28
Our agent loop · park_agent.pypython
# Our park agent: a tool loop with a checked answer (Modern Age Coders, written for this page)
import csv

PARKS = list(csv.DictReader(open("ottawa-parks.csv", encoding="utf-8-sig")))
DOG_CODES = {"off_leash": "0", "on_leash": "1", "mixed": "2", "no_dogs": "3", "not_designated": "4"}

def find_parks(ward, dog_rule):
    """Tool 1: park names in a ward with a given dog rule."""
    code = DOG_CODES[dog_rule]
    return [p["NAME"] for p in PARKS if p["WARD_NAME"] == ward and p["DOG_DESIGNATION"] == code]

def park_details(name):
    """Tool 2: address and the city's own dog wording for one park."""
    for p in PARKS:
        if p["NAME"] == name:
            return {"address": p["ADDRESS"], "dogs": p["DOG_DESIGNATION_DETAILS"]}
    return {"error": "no park called " + name}

TOOLS = {"find_parks": find_parks, "park_details": park_details}

def run(model_steps):
    """The loop: run each tool call the model asks for, keep every result, check the final answer."""
    seen = set()
    for step in model_steps:
        if step["type"] == "tool":
            result = TOOLS[step["name"]](**step["args"])
            names = result if isinstance(result, list) else [step["args"].get("name", "")]
            seen.update(names)
            print("tool", step["name"], "->", len(names) if isinstance(result, list) else result.get("address", result))
        else:
            unsupported = [n for n in step["parks"] if n not in seen]
            print("answer:", step["parks"][:3], "..." if len(step["parks"]) > 3 else "")
            print("check:", "OK" if not unsupported else "REJECTED, no tool result for " + ", ".join(unsupported))

# Stand-in for a model: a fixed plan, so the loop can be tested without any AI service.
found = find_parks("Kanata North", "off_leash")
honest = [{"type": "tool", "name": "find_parks", "args": {"ward": "Kanata North", "dog_rule": "off_leash"}},
          {"type": "tool", "name": "park_details", "args": {"name": found[0]}},
          {"type": "answer", "parks": found}]
invented = honest[:1] + [{"type": "answer", "parks": found[:2] + ["Maple Leaf Dog Run"]}]

run(honest)
run(invented)
Output on the city's filetext
tool find_parks -> 11
tool park_details -> 5599 Campeau Drive, Kanata
answer: ['Kanata Town Centre Stormwater Pond', 'Judy Laughton Park', 'Village Green Stormwater Facility'] ...
check: OK
tool find_parks -> 11
answer: ['Kanata Town Centre Stormwater Pond', 'Judy Laughton Park', 'Maple Leaf Dog Run']
check: REJECTED, no tool result for Maple Leaf Dog Run

Kanata North has 95 parks; 11 carry code 0. The honest plan passes the check. The plan that adds a park its tools never returned is rejected.

Why the stand-in, and what replaces it

The model is replaced here by a fixed plan, clearly marked, so the loop and its check can be run and tested on any laptop without an AI service. That is how we test agent code in class: the plan that a model would produce is written out by hand first, including a bad one.

In the adult course the fixed plan is swapped for calls to a model through its API, and nothing else changes: the same tools, the same loop, the same check. For teens, the same exercise is done with tools their age allows.

Three weaknesses we leave in on purpose

park_details returns the first park with a matching name, but 16 names appear more than once in the file. A better tool takes the PARK_ID instead. Learners find this themselves by asking for a name that repeats.

The check only proves each named park came from a tool; it cannot prove the agent asked the right question. And the meaning of codes 0 to 4 lives in our DOG_CODES table, which we wrote by reading the city's text. If the city changed its codes, the agent would be confidently wrong. Each of these is a class discussion.

Data: City of Ottawa open data, Parks and Greenspace. Contains information licensed under the Open Government Licence, City of Ottawa. The licence allows use "including for commercial purposes". Code, tallies and the stand-in plan are ours; the City of Ottawa does not endorse this page.

In Canadian classrooms

Where AI-assisted coding already appears in Canada

Canadian school documents have started to name AI tools and to ask students to judge them. These are the lines we read on 22 October 2026.

AI in Canadian course and contest documents (read 22 October 2026)
WhereWhat it saysWhat it means for a learner
New Brunswick, Computer Science 110Students code "using coding environments or AI-assisted debugging tools"A provincial course already expects AI help in debugging
Ontario, ICD2O (2023)Students investigate innovations "including automation and artificial intelligence, and assess their benefits and limitations"Judging AI is part of the course, not just using it
Canadian Computing Competition rulesGenerative AI forbidden; "AI-based tools within an IDE cannot be used."Contest skill has to be the student's own

For teens in school

These documents pull in two directions, and both are right. A teen should be able to use an allowed assistant to find a bug, and should also be able to write the program with no assistant at all, because a contest, a test or an interview will ask for exactly that.

Our teen classes practise both, in the same week.

For adults at work

For adults the question is different: how to use agents without losing control of the code. Version control, small reviewable changes and tests that run before anything is merged are the habits that make agents safe to use on real work.

We make no job or salary promises. What we teach is the judgement that keeps AI-written code from becoming someone else's problem.

Thinking, not copying

Why we teach vibe coding as a reading skill

Our promise is the same in every country: learners are taught to think a problem through and solve it, not to copy an answer or an AI's output. With vibe coding that promise is tested hardest, so we make it concrete.

Predict before accepting

Before running AI-written code, the learner says what it should do and where it might fail. Accepting code you could not predict is how bugs reach users.

Explain every line

Anything a learner submits, they can explain line by line. A line they cannot explain is rewritten by them, by hand.

Write the check first

For agents, the test or the answer check is written before the agent runs, exactly as in the Ottawa example.

The Collins definition describes vibe coding as "telling a machine what you want rather than painstakingly coding it yourself". Collins adds: "It's programming by vibes, not variables." That is a fair description of the first step. Our classes are about every step after it, where the variables come back.

Six zones, one plan

Class times for teens and working adults across Canada

Teens fit classes around school; adults around work and family. Every class is booked in the learner's own zone, and only the eastern provinces change clocks this year.

UTC offsets, 2026 to 2027 school and work year (IANA tz database 2026e)
WhereUntil 14 March 2027From 14 March 2027
British Columbia (most) and YukonUTC-7UTC-7
Alberta and SaskatchewanUTC-6UTC-6
ManitobaUTC-5UTC-5
Ontario and QuebecUTC-5 (from 1 November 2026)UTC-4
Maritime provincesUTC-4 (from 1 November 2026)UTC-3
NewfoundlandUTC-3:30 (from 1 November 2026)UTC-2:30

Adults in our courses often choose weekday evenings or weekend mornings. Our teachers are in India, where a Canadian weekend morning falls in the evening, so those slots are easy to arrange in every zone.

Four levels

From prompting to building agents you can trust

Our levels for placing a teen or adult in the first class.

Four levels in vibe coding and agents (ours)
LevelWhat the learner can do
1. PromptsGets working code from an assistant for a small task
2. ReadsExplains AI-written code line by line and fixes its mistakes
3. Builds toolsWrites functions with clear inputs, outputs and errors for an agent to call
4. Checks agentsWraps an agent in tests and answer checks, and knows when not to use one

For parents of teens

Level 2 is the one that matters most at school age. A teen who can read code can learn anything that follows.

Teens reach level 3 in our course with tools their age allows; agentic tools for adults come later, at 18.

For adults

Most self-taught vibe coders stop at level 1. Levels 2 and 4 are where AI-written code becomes something you can put your name to.

Keep every agent change in version control. If you cannot undo it, you cannot trust it.

Related courses

Nine more courses around AI and building

Live, with one regular teacher, in a batch or one to one. Teen and adult courses are kept separate.

I

Teens

Ages 13 to 18

VAAC / TEEN / 01

Python for teens

The language most AI tools write.

Open the syllabus

VAAC / TEEN / 02

Git and GitHub for teens

Undo anything an assistant changes.

Open the syllabus

VAAC / TEEN / 03

AI and machine learning for teens

How the models behind the tools work.

Open the syllabus
II

Adults: building with AI

Ages 18 and up

VAAC / ADULT / 01

Generative AI

LLMs, retrieval and agents, from first principles.

Open the syllabus

VAAC / ADULT / 02

Python automation

Scripts and applied AI for real work.

Open the syllabus

VAAC / ADULT / 03

Full stack development

React, Node.js and deployment basics.

Open the syllabus
III

Adults: agents at work

Professionals

VAAC / PRO / 01

Gemini Enterprise agents

Agents in a Google workplace.

Open the syllabus

VAAC / PRO / 02

Copilot Studio agents

Agents in a Microsoft workplace.

Open the syllabus

VAAC / PRO / 03

Advanced Git and GitHub

Review and control at team scale.

Open the syllabus

How classes run

Live classes, teens and adults apart

Weekdays or weekends, flexible: choose your batch, or for a very personalised plan choose 1-on-1 at your convenience, weekend mornings or weekdays.

Weekdays

After school for teens; evenings for adults.

Weekends

Mornings or afternoons, in your own zone.

1-on-1

Built around your project or your codebase.

Age-correct tools

Teens and adults use what their age allows.

Read, then accept

No AI-written line is kept unexplained.

Tools before agents

Reliable functions first, then the loop.

Checks on answers

Every agent answer is compared with its tools.

Version control

Every change can be inspected and undone.

Group or solo

A batch of fifteen to twenty, or one to one.

Student work

Projects our students have shipped

Some of these were built with AI assistance and then read, tested and fixed by the student. More are on the student labs page.

NutriLife AI nutrition coach project screenshot

AI and ML

NutriLife

An AI nutrition coach that reads what you eat and works you toward a target.

by Bhavya · Open it

Misti AI chatbot for maths and coding screenshot

AI and ML

Misti

A chatbot that answers mathematics and programming questions, built and deployed by a student.

by Harshit · Open it

GuardianX AI internet safety assistant screenshot

AI and ML

GuardianX

An assistant that helps a young person recognise unsafe situations online.

by Vivaan · Open it

SkyCast weather forecast application screenshot

Web app

SkyCast

A weather forecasting site with live conditions for any location.

by Krish

Fees

Fees

Courses are paid by the month, and the fee covers eight live classes. Adults fitting classes around work can pause between months without anything owed.

Free first class

Free

no card required

  • A free first class
  • A short read-the-AI-code exercise
  • No card details
Book it

Group batch

$140 CAD

a month, eight live classes

  • Fifteen to twenty learners of one age group
  • A regular teacher
  • A weekly slot that suits you
  • Billed monthly
Start here

One to one

$210 CAD

a month, eight live classes

  • A teacher to yourself
  • Your own project or repository
  • Times around your week
Enquire

What families say

Rated 4.9 across 547 Google reviews

Reviews on Google from students and parents, as they wrote them.

★★★★★

"The one step solution for my son. Modern Age Coders make learning coding so simple that kids love it. The teachers explain complex concepts clearly with practical exercises and interactive content."

Ria Mukherjee

Parent

★★★★★

"Modern Age Coders has been a game-changer for me. I struggled to grasp IT concepts and coding before joining, but their classes transformed everything. I can now confidently write complex programs with ease."

Samriddha Mondal

Student

★★★★★

"One of the most wonderful education centres out there. Education is not limited to school syllabus but focuses on skill development."

Vansh Agarwal

Student

★★★★★

"My child Dhairya is really enjoying the Modern Age Coders classes. This is his first online class and he eagerly looks forward to it. I can already see his improvement, and the teachers are very cooperative."

Sonam Oswal

Parent of Dhairya

★★★★★

"Modern Age Coders have wonderful teachers who teach in a clear, easy and practical way. The teacher boosts students' confidence and inspires them to learn without hesitation."

Sonu Goyal

Parent

★★★★★

"I highly recommend this computer coding class! The teachers are incredibly knowledgeable and passionate about coding."

Ritu Kedia

Parent

Questions about vibe coding and agents

Vibe coding and AI agents in Canada: common questions

What is vibe coding?

Collins Dictionary, which made it its Word of the Year 2025, says vibe coding refers to the use of artificial intelligence prompted by natural language to write computer code. In our classes it is the first step; reading and testing what the AI wrote is the rest.

What is the difference between an AI agent and a workflow?

In Anthropic's definitions, a workflow runs models and tools through predefined code paths, while an agent lets the model direct its own process and tool use. Agents are more flexible and also more prone to compounding errors.

Can teenagers take vibe coding classes?

Yes. Our Vibe Coding for Teens course is for ages 13 to 17 and uses only tools whose terms allow the student's age, such as GitHub Copilot for verified students and, with a parent's permission, OpenAI's tools.

Can a teen use Claude or Claude Code?

No. Anthropic's terms say users must be at least 18 years old. We never recommend Claude or Claude Code to anyone under 18; they appear only in our adult courses.

Is there an AI agents course for adults in Canada?

Yes. Our Codex and Claude Code course for professionals runs 24 live classes over 12 weeks, from agent foundations to securing and shipping. Vibe Coding for College covers building full stack projects with AI assistance for ages 18 and up.

Do I need to know how to code first?

It helps but is not required. Every course starts by teaching learners to read code, because judging AI-written code is the core skill. Adults new to programming often take a Python course alongside.

Can AI agents replace programmers?

We make no such claim. Anthropic's own guidance warns of higher costs and compounding errors and recommends testing agents in sandboxed environments. Our own small agent shows an answer naming a park its tools never returned; only a check caught it.

Can students use AI tools in the Canadian Computing Competition?

No. The CCC rules forbid generative AI and AI-based tools inside an editor during the contest, so contest practice happens without assistants.

What times are classes in Canada?

Weekdays or weekends, flexible: choose your batch, or for a very personalised plan choose 1-on-1 at your convenience, weekend mornings or weekdays. Every time is set in your own province's zone.

What will I build?

Teens build Python and web projects with allowed assistants and explain every line. Adults build agents on real codebases and data, such as the Ottawa parks agent on this page, wrapped in tests and answer checks.

Elsewhere on this site

Related pages

More on building with AI, and more for learners in Canada.

Coding classes in Canada

Our pages for every province.

AI and machine learning in Canada

Training models, not just prompting them.

What is vibe coding?

The idea in more depth.

Vibe coding for teens

Our teen approach in detail.

A parent's guide to vibe coding

For parents of teens.

AI agents explained

How agents decide what to do.

Agentic coding classes

Working with coding agents.

Model Context Protocol course

Connecting agents to tools.

AI tools age guide

Age rules for many tools.

Start here

Book a free first class

We start with a short piece of AI-written code that has a hidden mistake. Finding and explaining it tells the teacher your level, and whether the teen or adult course fits.

Want to read more? How we teach explains a live class, and the course pages set out each syllabus.

WhatsApp us · +91 91233 66161 · connect@modernagecoders.com

Classes are live over video, with teachers based in India, for learners in every province and territory. The fastest way to reach us is WhatsApp.

The first class is free and needs no card.

Want a class sooner? Book a priority demo

A priority demo is a full class of about 45 to 60 minutes. After the class the mentor sends a written skill report and a personal learning roadmap on WhatsApp. If you enrol, the demo fee is adjusted against your first month's fee. Share your demo Order ID with us on WhatsApp and we take it off.

Book a priority demo

Keep exploring Modern Age Coders

Coding classes in nearby places

Learn more

Free resources

From the blog

Start here

Free classPriority demoCall us