Interactive learning material · M1-T2-MLC1

Managing Complex Tasks with AI

ChatGPT Canvas, Projects, Codex

Author: Zsolt LengyeliTStudy Hungary Training and Research Centre
Reviewer: Mária HartyányiiTStudy Hungary Training and Research Centre

Instructions, documents and conversations are brought together in a shared project folder — A structured workspace for extended professional tasks

Learning objective

Planning a teaching unit, developing a course, building a question bank or preparing project documentation are not one-prompt tasks. Advanced AI environments can serve as structured workspaces for these extended professional workflows.

By the end of this material, you will be able to

break a complex task into manageable stages with human checkpoints;
use Projects to keep files, instructions and conversations together;
edit longer texts and code iteratively in writing and code blocks;
understand what an agent such as Codex can do;
keep control over aims, quality and final decisions throughout.

From one-off prompts to workflows

When a single request is no longer enough

Most people start using AI with isolated questions: one prompt, one answer. This works well for short tasks, but longer work, tasks with several parts or work that continues over days need a different approach.

The conversation can grow too long, earlier decisions may be forgotten or compressed, and files become scattered across chats. It becomes unclear which version is the latest.

Environments designed for longer work help keep related material together, revise documents step by step and carry out multi-step tasks.

Project spaces

Keep related files, instructions and conversations together.

Editors

Allow you to revise a document or code step by step with targeted changes.

Agents

Plan and execute steps on the computer within the permissions you provide.

A staged workflow

Five stages with human review

The most important technique is not tool-specific: split the work into stages and insert human checkpoints between them.

Five stages: plan, draft, revise, check and finalise, with human checkpoints between them — Plan → draft → revise → check → finalise

Plan
Clarify aims, audience, scope and criteria. Ask the AI to propose a structure and ask you clarifying questions.
Draft
Generate the parts one by one. Each new part should build on the approved plan.
Revise
Improve specific sections with targeted instructions: “Rewrite only section 3 for a lower reading level.”
Check
Verify facts, safety content, learner level and alignment with the curriculum. Ask the AI to critique its own draft against your criteria, then check it yourself.
Finalise
Format, export and store the final version together with the key prompts that produced it.

Stages in practice

A teaching unit, step by step

Stage 1 · Plan

Start with the plan

“Here is the curriculum excerpt (attached). Propose a 4-week unit plan for second-year HVAC apprentices: weekly aims, activities, assessment. Ask me questions first.”

Checkpoint: the teacher approves or modifies the plan.

Stage 2 · Draft

One part at a time

“Using the approved plan, draft the worksheet for week 1 only.”

This lets you review the worksheet on its own before moving on to the next week.

Stage 3 · Revise

Targeted revisions

Adjust the difficulty level, wording or a specific safety step.

Checkpoint: the teacher carries out a technical and safety check of the worksheet.

Stage 4 · Question bank

Build on approved material

“Create 10 quiz questions for weeks 1–2 with answer key, based only on the approved worksheets.”

After checking the questions and answer key, you can finalise the material.

Projects

A workspace for related work

In ChatGPT, a Project is a smart folder that keeps related chats, files and instructions together, providing consistent context for the work within the project.

Similar functions exist in other assistants, such as Projects in Claude or Gems and notebooks in Google’s tools.

Project instructions tell the assistant how to behave within this project only. They take precedence over general custom instructions.

A Project keeps instructions, reference files and separate conversations in a shared folder — Shared instructions and files, focused conversations

Example project instructions

“You help me develop materials for the automotive mechatronics course. Always use British English terminology, include safety notes, and produce tables with answer keys.”

Project settings

Files, memory and sharing

Reference files

The syllabus, assessment criteria and style guide are available to every chat in the project.

The number of files depends on the plan. At the time the source document was written: 5 in Free; 25 in Go/Plus; 40 in Edu, Pro, Business and Enterprise.

Project memory

Memory can be set to default or project-only.

With project-only memory, chats cannot draw on conversations outside the project. This is useful for keeping courses or confidential work separate.

Joint development

You can share projects with colleagues, who can chat in or edit the project depending on their role.

Shared projects automatically use project-only memory.

The plan limits shown here are from the 2026 source document. For current details, see the Projects help article linked in the Sources chapter.

Editing text and code

From Canvas to writing and code blocks

In 2024, OpenAI introduced Canvas, a side-panel editor in which the user and ChatGPT could work on the same document or code instead of regenerating whole answers.

According to the source document, Canvas was withdrawn from GPT-5.5 models in May 2026; its functions now appear directly in the chat as writing and code blocks.

Other assistants offer comparable workspaces, such as Artifacts in Claude or Canvas in Gemini.

A conversation alongside editable text with a selected section and a runnable code block — Targeted text editing and code handling within the conversation

Writing blocks

Editing longer texts

You can open longer texts in a focused full-screen editor, edit them, copy them and download them.

Code blocks

Preview and run

Code blocks can preview or run their content. You can use them for small interactive elements, such as a self-check quiz or a calculator for a workshop formula.

The editing method

One change at a time

Whatever the tool is called, the working method is the same: ask for a first draft, then edit it in place, either yourself or with targeted requests for selected parts.

1. First draft

Establish a shared starting point

Ask for a draft based on the approved plan. Continue working on the existing version rather than regenerating the whole document.

2. Targeted change

Name one specific task

“Shorten the introduction.”

“Add a safety note after step 4.”

This makes the effect of the change visible and easy to check.

3. Version keeping

Save the approved version

Download or copy the approved version before asking for major changes, so that you can return to it later.

4. Interactive elements

Check before use

Use a code block for a small quiz or calculator. Test the completed element before using it in a lesson.

Codex

An agent that works on the computer

Codex is OpenAI’s AI agent originally built for writing, reviewing and running code. The source document describes access through ChatGPT plans, with plan-dependent usage limits, in the desktop app, on the web, in the command line and in code editors.

An agent plans steps and executes them: it reads and writes files, runs programs and, with the tools available to it, can use a browser or other applications.

Comparable agents include Claude Code and Claude Cowork. You do not need to be a programmer to benefit from them.

Teaching material development

Quizzes with scoring

Creating questions, answer keys and interactive self-check exercises.

File management

Batch operations

Renaming or reformatting dozens of files according to a consistent rule.

Classroom support

Small teaching tools

Building a calculator or another simple application that supports a lesson task.

Directing the agent

Clear instructions and appropriate permissions

Because an agent acts, choose consciously how much freedom to give it. Start with a mode in which it asks for approval, observe how it works, and only then give it more freedom.

For important material, let it work in a separate folder with copies of your files. Keep the originals; never let it work on the only copy.

State clearly whether you want exploration or action. If you only want an analysis, say so explicitly. Watch its reasoning while it works; if it goes in the wrong direction, stop it and correct the instruction.

Three permission levels allowing progressively more file operations, program execution and network access — The user decides how much freedom the agent has

Asks for approval

The agent asks before steps that require permission.

Proposes changes for review

You can inspect the proposed changes and decide whether to accept them.

Broader access

The agent has more freedom to carry out the task. Choose this consciously and according to the task’s needs.

Keeping the teacher in control

Checklist for extended AI workflows

Aims and criteria. Define aims, audience and quality criteria before you start, and write them into the project instructions.
Checkpoints. Approve the plan before drafting, and check facts and safety content before finalising.
Versions and prompts. Keep the originals and approved versions; record the key prompts.
Data protection. Never upload learners’ personal data. Use institutional accounts where available.
Professional responsibility. The final decision and professional responsibility remain with the teacher.

Practical application

Try it with your own course unit

Individual work · 40 minutes

My course unit as a Project

Create a Project for a real course unit. Write 5–8 lines of project instructions covering the subject, learner level, terminology, format expectations and safety requirements.

Upload 2–3 reference files without personal data.
In a separate chat, ask for a plan and approve or modify it.
In a new chat, ask for a worksheet draft. Make at least two targeted edits.
Check facts and safety content yourself. Ask the AI to evaluate the worksheet too, then download the final version.

Group work · 45 minutes · 3–4 learners

AI-supported mini project

Learners produce a short workshop guide, such as a maintenance checklist. Assign the roles of planner, AI operator and checker.

Plan: ask for a structure and record what the group decides to keep.
Draft: have the AI draft one section at a time; the group edits it in its own words.
Check: compare the draft with the textbook or workshop rules.
Present the guide and the decision log.

If learners cannot use their own AI accounts, the teacher can take the AI operator role on the projector. Evaluate the process as well as the final product.

Download the detailed practical exercisesM1-T2-MLC1-EXC-EN.pdf

Takeaways for teachers

Takeaways for your next task

Think in workflows. Break the complex task into stages and review between stages.
Use one Project per course or unit. Provide shared instructions and reference files; start new sub-tasks in new chats.
Edit iteratively. Change one thing at a time in longer texts and code, and keep approved versions.
Direct the agent. Give agents such as Codex clear instructions, limited permissions and copies of files.
Keep the final decision. AI accelerates the work; aims, quality and final decisions stay with you.

Self-check

Check your understanding!

There is one correct answer to each question. Answer all seven questions, then check your answers.

1What is the key principle for managing a complex AI task?Include every detail in one long prompt.Break the work into stages with human checkpoints.Leave all review to the AI.

2What is a Project for?Keeping related conversations, files and instructions together.Storing only final documents.Automatically replacing the teacher’s decisions.

3What does project-only memory mean?The project forgets every earlier conversation.It automatically uses conversations from every other course.Chats cannot draw on conversations outside the project.

4How should you revise an already approved text?Regenerate the whole material for every small correction.Make one targeted change at a time and keep the approved version.Delete all saved versions before editing.

5What characterises an agent such as Codex?It plans and executes steps, such as handling files or running programs.It can only give text answers and cannot work with files.It independently takes over the teacher’s professional responsibility.

6You only want the agent to analyse. Which instruction is the clearest?“Take care of everything in the folder.”“Change the files as you see fit.”“Analyse the documents and make recommendations; do not modify the files.”

7What should the teacher do before finalising?Accept the AI’s self-evaluation without checking it.Personally check facts and safety content too.Upload learners’ personal data for better personalisation.

References and links

Sources and further reading

References in the source document

OpenAI Help Center (2026): Projects in ChatGPT; Using Codex with your ChatGPT plan; ChatGPT release notes (28 May 2026: Canvas replaced by writing and code blocks).
OpenAI (2026): Introducing the Codex app.
Lengyel, Zs. (2026): Introduction to the concepts of artificial intelligence; Concrete applications for content generation. VITA training presentations.

Project spaces

Projects in ChatGPT

Instructions, files, project memory and shared work.

Open the Projects help article

Agents

Using Codex

Access and use with a ChatGPT plan.

Open the Codex help article

Product updates

ChatGPT release notes

Changes to features and editing options.

Open the release notes

Closing page

Credits and copyright

Author: Zsolt LengyeliTStudy Hungary Training and Research Centre
Reviewer: Mária HartyányiiTStudy Hungary Training and Research Centre
Project identifier: 2025-1-LT01-KA220-VET-000362456

Attachments

Download the English learning materialM1-T2-MLC1-TEXT-EN.pdf Download the practical exercisesM1-T2-MLC1-EXC-EN.pdf

Creative Commons Attribution-ShareAlike licence

This work may be reused and adapted under the terms of the Creative Commons Attribution-ShareAlike 4.0 International License (CC BY-SA 4.0).

Funded by the European Union. Views and opinions expressed are those of the author(s) only and do not necessarily reflect those of the European Union or the National Agency.