教育のライフサイクルを支える AI エージェント入門：学校現場での設定から活用まで

はじめに｜Introduction

🇯🇵 日本語

「AI を授業に使ってみたい」と思っている先生は増えています。しかし、ChatGPT にひとつ質問して終わり、というような使い方では、AI の本当の力はまだ引き出せていません。

AI エージェント（Agentic AI） は、単に質問に答えるツールではありません。目標を与えると、複数のステップを自律的にこなし、ツールを使い、自分の出力を検証しながら、完成した成果物を届けてくれるシステムです。

この記事では、授業設計・授業実施・評価・振り返りという「教えることのライフサイクル」全体にわたって、AI エージェントをどう設定し、どう活用するかを、ステップ・バイ・ステップで解説します。対象読者は、AI に興味はあるけれど何から始めればいいかわからない学校の先生・教育管理職・カリキュラム担当者です。

この記事を読んでわかること：

• AI エージェントと通常の AI ツールの違い
• 教育現場でのエージェント設定の具体的な手順
• 小中高での実際のユースケース（Year 9 英語・IB English A・数学・理科・学校管理）
• 責任ある使い方のための原則

🇬🇧 English

More teachers are curious about using AI in the classroom — but asking ChatGPT a single question and moving on barely scratches the surface of what's possible.

Agentic AI is not simply a question-answering tool. Give it a goal and it works through multiple steps autonomously, uses external tools, checks its own outputs, and delivers a finished product for you to review.

This article walks you through — step by step — how to set up and use AI agents across the full lifecycle of teaching: planning, delivery, assessment, and reflection. It is written for teachers, department heads, and curriculum coordinators who are interested in AI but unsure where to start.

What you'll learn:

• The difference between agentic AI and standard AI tools
• Concrete steps to configure an agent for your classroom context
• Real use cases across school settings (Year 9 English, IB English A, maths, science, school administration)
• Principles for responsible, teacher-led use

AI エージェントとは何か｜What Is an AI Agent?

🇯🇵 日本語

通常の AI ツール（例：ChatGPT）は、あなたが入力した質問に対して 1 回答えて終わります。次の一手はまたあなたが考える必要があります。

AI エージェント は、それとは根本的に異なります。

通常の AI ツール	AI エージェント
動き方
1 問 1 答	目標に向けて複数ステップを自律実行
記憶
会話の中だけ	長期記憶・外部データへのアクセスが可能
ツール使用
なし（または限定的）	検索・ファイル読み込み・カレンダー連携など
アウトプット
テキスト回答	完成した成果物（単元計画・評価レポートなど）
教師の役割
毎ステップで指示	最初に目標を与え、最後にレビュー

教育の文脈でいえば、エージェントは「とても優秀で疲れを知らない TA（ティーチング・アシスタント）」のようなものです。授業計画の草案を書き、差別化された教材を生成し、フォーマティブ評価を設計し、採点のサポートをし、その結果を次の授業設計にフィードバックします。そして、すべての最終判断は先生が行います。

🇬🇧 English

A standard AI tool (like ChatGPT) answers your question once and waits. What comes next is up to you.

An AI agent is fundamentally different. It accepts a goal, then works through the steps needed to reach it — gathering information, generating content, checking its own work, and delivering a finished output.

In an educational context, think of the agent as a highly capable, inexhaustible teaching assistant. It drafts unit plans, generates differentiated materials, designs formative assessments, supports marking, and feeds results back into the next planning cycle. The teacher makes every final decision.

教育のライフサイクルと AI エージェント｜The Teaching Lifecycle & AI Agents

🇯🇵 日本語

教えることには、繰り返されるサイクルがあります。

計画（Plan）→ 準備（Prepare）→ 実施（Deliver）→ 評価（Assess）→ 振り返り（Reflect）→ 計画（Plan）…

AI エージェントは、このサイクルのすべてのフェーズでサポートできます。

🇬🇧 English

Teaching follows a repeatable cycle:

Plan → Prepare → Deliver → Assess → Reflect → Plan again…

AI agents can support every phase of this cycle — not just lesson planning, but the whole arc from curriculum design through to end-of-year reflection.

ステップ・バイ・ステップ：エージェントのセットアップ｜Step-by-Step: Setting Up Your Agent

🇯🇵 日本語

以下の手順は、Claude（Anthropic）、ChatGPT（OpenAI）、Gemini（Google）など、主要なエージェント対応 AI プラットフォームで共通して使える考え方です。ツール固有の UI 操作は異なりますが、考え方と入力する内容は共通です。

🇬🇧 English

The steps below apply to any major agentic AI platform — Claude (Anthropic), ChatGPT (OpenAI), Gemini (Google), and others. The specific UI differs across tools, but the logic and the information you provide are universal.

ステップ 1：常設コンテキストを設定する｜Step 1: Configure Your Standing Context

🇯🇵 日本語

エージェントに最初に与える「常設コンテキスト」は、以降のすべての作業の基盤になります。これは一度設定すれば、毎回入力し直す必要はありません。

設定に含めるべき情報：

# 教師プロフィール
- 担当教科：英語（First Language English）
- 学年：Year 9（中学 3 年相当）
...

🇬🇧 English

The standing context is the foundation the agent draws on for every task. Set it once; don't re-enter it each time.

What to include:

# Teacher profile
- Subject: English (First Language English)
- Year group: Year 9
...

ステップ 2：単元ブリーフを渡す｜Step 2: Give the Unit Brief

🇯🇵 日本語

単元ごとに、エージェントに具体的な目標と制約を与えます。これが「タスクブリーフ」です。

例：Year 9 説得力のある文章を書く単元

## タスクブリーフ
目標：Year 9 クラス向けに、説得力のある文章（オピニオンピース）を書く
10 コマの単元を設計してほしい。
...

このブリーフを送ると、エージェントは次のことを自律的に行います（30〜60 秒）。

🇬🇧 English

For each new unit, give the agent a focused brief. This takes 3–5 minutes to write.

Example: Year 9 persuasive writing unit

## Unit brief
Goal: Design a 10-lesson unit on persuasive writing
(opinion pieces) for my Year 9 class.
...

Once you submit this brief, the agent works through the following steps autonomously (30–60 seconds):

ステップ 3：エージェントが自律的に行うこと｜Step 3: What the Agent Does Autonomously

🇯🇵 日本語

① Standing context (curriculum, class profile, rubric) 를 읽기
↓
② 가르쳐야 할 기술을 식별하고 기초에서 발전으로 순서立て기
...

🇬🇧 English

① Read standing context (curriculum, class profile, rubric)
↓
② Identify skills to teach; sequence from foundational to extended
...

ステップ 4：先生によるレビューと編集｜Step 4: Teacher Review and Editing

🇯🇵 日本語

Agent 의 출력은 반드시 초안입니다. 최종 결과물ではありません.

리뷰 체크리스트:

• 전체 유닛의 흐름이 논리적인가
• 멘토 텍스트는 자신의 클래스에 맞는지 (문화적 문맥, 학생 관심사)
• 페이싱은 자신의 클래스의 리듬과 일치하는가
• 차별화 작업은 개별 학생의ニーズ를 정말로 반영하고 있는가
• 학생용 자료의 톤앤매너는 자신의 것이 되는가
• 에이전트가 가정했던 사항 중 수정이 필요한 것은 있는가

Typical editing time: 20–30 minutes (Agent 가 몇 시간 걸리는 작업을 대체한 후)

🇬🇧 English

The agent's output is always a draft. Review it before using any part of it.

Review checklist:

• Does the overall unit arc make sense?
• Are the mentor texts right for your class (cultural context, student interests)?
• Does the pacing match your class's rhythm?
• Do the differentiated tasks genuinely reflect individual student needs?
• Does the tone of student-facing materials sound like you?
• Are there any assumptions the agent made that need correcting?

Typical editing time: 20–30 minutes (after the agent has replaced hours of blank-page work)

ステップ 5：フォローアップ・プロンプトで反復する｜Step 5: Iterate with Follow-Up Prompts

🇯🇵 日本語

첫 번째 초안을 확인한 후, 관심 있는 부분을 추가 프롬프트로 수정합니다.

Follow-up prompt 예시:

「7 수업시간을 더 토론 중심으로 바꾸고,
교사 주도 시간을 줄이세요。」
「EAL 학생용 Task 4 의 스캐폴딩 버전은
...

🇬🇧 English

After reviewing the first draft, refine specific elements with follow-up prompts. Each one takes seconds to write.

Example follow-up prompts:

"Make lesson 7 more discussion-based and less teacher-led."
"The scaffolded version of Task 4 is still too complex
for my EAL students — simplify it further."
...

ステップ 6：結果をフィードバックして改善する｜Step 6: Feed Results Back In

🇯🇵 日本語

유닛 종료 후, 평가 데이터와 반성을 에이전트에게 되돌려줌으로써 다음 유닛 계획이 더 똑똑해집니다.

## 유닛 후 피드백
총괄 과제 결과:
- 클래스 평균: 72%
...

이를 통해 에이전트는 개정된 유닛 계획과 타겟팅된 보충 활동을 생성합니다.

🇬🇧 English

At the end of each unit, feed assessment data and reflective notes back to the agent. The next unit plan will be smarter for it.

## Post-unit feedback
Summative results:
- Class average: 72%
...

The agent returns a revised unit plan and targeted supplementary activities.

ユースケース：学校現場での実例｜Use Cases: Real School Scenarios

🇯🇵 日本語

다음에, 다양한 과목 및 상황에서 AI 에이전트 활용 예시를 보여줍니다.

🇬🇧 English

Here are concrete examples of how AI agents can be used across different subjects and school roles.

Use Case 1: Year 9 English — Unit Planning & Differentiation

Scenario: A Year 9 English teacher (26 students, mixed ability, 4 lessons per week) needs to design a persuasive writing unit. She has 45 minutes of prep time available.

What she gives the agent:

• Class profile (4 EAL, 1 LSP, 5 high-achieving)
• A 10-lesson unit aimed at a summative task in two weeks
• Cambridge IGCSE assessment criteria

What the agent produces (~40 seconds):

• 10-lesson sequence with daily objectives
• 3 versions of each task (standard / scaffolded / extended)
• 12 mentor texts with annotation guides
• 5 exit tickets (one per 2 lessons)
• Peer feedback protocol

Teacher editing time: 25 minutes — mainly swapping mentor texts and adjusting pacing.

Sample prompt used (click to expand)

Design a 10-lesson unit on persuasive writing (opinion pieces)
for my Year 9 First Language English class.
Summative task: a 600–800 word opinion piece on a topic of the
...

Use Case 2: IB English A — Exam Revision & Individual Focus Cards

Scenario: An IB DP2 teacher (14 students) has 8 weeks until IB exams. Mock results show class-wide weaknesses in Paper 1 (unseen text analysis) and Paper 2 (comparative essay structure).

What she gives the agent:

• Mock exam scores and her own handwritten observations per student
• IB English A HL criteria (uploaded PDF)
• 2 lessons per week × 6 weeks

What the agent produces:

Use Case 4: Science — Inquiry-Based Unit Design

Scenario: A Year 8 science teacher wants to design an inquiry-based learning unit on "Ecosystems and Human Impact". They want to include IB ATL skills (inquiry, critical thinking, communication).

What she gives the agent:

• Unit theme and IB MYP Science criteria
• Requirements for mapping ATL skills
• Constraints: 3 weeks, 12 periods
• Lab resources (microscopes and data loggers available, no field trips)

What the agent produces:

• A unit framework centered around an inquiry question
• Activities organized by phase (Introduction → Inquiry → Analysis → Presentation)
• Three inquiry paths for students to choose from (water quality, soil health, urban heat)
• Scaffolding for groups and resources
• Draft rubrics integrating ATL skills into assessment
• Checkpoints for each phase (for both teacher and student use)