⚑ AI Routine · Agent Framework

How I Use AI Routines
in My Computer Class

By Andrew D.

A computer teacher teaching HTML does not need to manually write quizzes, check student code, or prep lessons every week. This page shows how a simple AI routine handles it, step by step, while the teacher stays in control.

⚑ Trigger Activation
+
🧠 Context Intelligence
+
πŸ›‘ Steering Alignment
=
🎯 Autonomous Value Real Output
Framework Breakdown
01 Β· Trigger
When does it run?
The event or schedule that starts the agent.
Schedule-Based
πŸ“…
Every Monday morning before class
Sends weekly lesson plan and objectives to students
πŸ“‹
End of each lesson (Friday)
Generates a recap summary and practice exercises
Event-Based
πŸ“€
Student submits HTML exercise
Agent reviews code and gives instant feedback
⚠️
Student scores below passing
Triggers extra practice set and a review tip
Custom Trigger
πŸ’¬
Teacher types a command
"Generate quiz for HTML forms" runs instantly
02 Β· Context
What does it know?
The knowledge and tools the agent works with.
Knowledge Sources
πŸ“š
Course syllabus and topic order
HTML basics, tags, links, forms, tables, CSS intro
πŸ—‚οΈ
Student progress records
Who is on which lesson, weak areas, past scores
πŸ›
Common beginner HTML mistakes
Unclosed tags, missing alt text, wrong nesting
Connectors
πŸŽ“
Google Classroom
Posts assignments and reads student submissions
πŸ“
Google Drive
Reads submitted HTML files from the student folder
πŸ“Š
Grading Sheet
Updates scores and flags after each submission review
Available Tools
✏️
Generate coding exercises by topic
Tags, forms, tables, images, links, semantic HTML
πŸ”
Check student HTML for errors
Flags issues with hints, not just the answer
πŸ“
Create quizzes with answer keys
Multiple choice, fill-in, and short code tasks
03 Β· Steering
How do you keep it honest?
Controls that keep the teacher in charge.
Human Oversight
πŸ‘€
Teacher reviews before sending
All exercises and quizzes go to teacher first
βœ…
Approve or edit outputs
Nothing reaches students without approval
Verification
πŸ§ͺ
Checks if generated HTML is valid
Agent tests its own output before showing it
πŸ“Š
Validates quiz answers are correct
Confirms answer key before distributing
Operational Controls
⏸️
Pause during exams or holidays
Teacher controls when the routine is active
↩️
Redirect to new topic anytime
Syllabus moves forward, agent follows
Live Scenario Β· HTML Forms Lesson
🏫
Computer Teacher Β· Web Programming Class
Topic: HTML Forms Β· Grade 10 Β· 25 students
1
Trigger Β· Every Sunday 8PM
Lesson prep starts automatically
The routine runs on schedule. It checks the syllabus, sees that Monday's topic is HTML Forms, and kicks off the preparation process with no manual input from the teacher.
2
Context Β· Connectors pull live data
Google Classroom and the grading sheet are read
Claude connects to Google Classroom to check last week's completion rates and reads the grading sheet to find which students scored below 70%. Google Drive is scanned for any submitted HTML files still pending review.
3
Context Β· Materials are generated
Quiz, exercise, and lesson summary ready in seconds
Using the syllabus and common mistake patterns, Claude creates a 10-question HTML Forms quiz with answer key, a hands-on coding exercise (build a contact form), and a short lesson summary for the board.
4
Steering Β· Teacher reviews before anything is sent
Approve, edit, or reject the draft
The teacher gets a clean draft to review. Adjust a question, reword the exercise instructions, or approve everything as-is. Nothing reaches students until the teacher says yes.
5
Trigger Β· Student submits HTML file
Submission lands in Google Drive, review starts instantly
When a student uploads their contact form HTML to the class Drive folder, the event triggers the agent. No need to check manually, it runs on its own.
6
Context Β· Code is checked
Agent reads the HTML and finds issues
Claude reviews the student's code for missing name attributes, wrong input types, unclosed tags. It writes feedback using hints, not direct answers, so students still have to think.
7
Steering Β· Score logged, teacher notified
Grading sheet updated, teacher reviews the summary
Score is written back to the grading sheet automatically. Teacher gets a one-line summary per student. Any submission flagged as needing attention is highlighted for a manual look.
8
Result Β· Autonomous Value
Teacher reviews for 10 minutes instead of working for 3 hours
Lesson prep, quiz creation, code review, and grade logging all happened automatically. The teacher focused on teaching, not paperwork. Students got feedback the same day they submitted.
Sample Prompt Sent to Claude
# ROLE You are a helpful assistant for a computer teacher. Your job is to prepare classroom materials for a web programming class. # TOPIC Current lesson : HTML Forms Grade level : Grade 10 Β· Web Programming Previous topic : HTML Links and Anchors (completed) # STUDENT CONTEXT Class size : 25 students Watch list : 3 students scored below 70% last week Common errors : missing name attributes, no labels, wrong input types # TASK 1. Generate a 10-question quiz Β· mix of multiple choice and fill-in 2. Create a coding exercise Β· build a contact form (name, email, message, submit) 3. Include an answer key Β· do not show this to students # TONE Clear, beginner-friendly, no technical jargon. Use hints in feedback, not direct answers.
Time Saved Β· Without vs. With Agent
Task Without Agent With Agent
Weekly lesson prep 60-90 min of manual typing Generated in seconds. Teacher reviews in 10 min.
Creating a quiz 30-45 min per quiz One command. Ready in under a minute.
Checking student HTML Review each file one by one Auto-flagged errors with hints per student.
Identifying struggling students Noticed after the next class Flagged immediately after low score.
Sending practice exercises Manually typed per student Auto-generated, personalized per student level.
Teacher control Full Β· but time-heavy Full Β· but time-efficient. Nothing sent without approval.
Want to build something like this?
I design and build AI-assisted workflows for real use cases. Let's talk.