From Chatbots to Agents

AI has moved from systems designed mainly to recognize patterns and make predictions, to tools that can generate new content and interact with us, and increasingly, to systems that can take action on our behalf.

To understand what that shift means, it helps to look at three broad waves of AI capabilities.

We’ve seen that many AI systems are designed to recognize patterns, sort information, or make predictions. Generative AI uses learned patterns to create new content.

During training, a generative model works through large amounts of data and learns patterns and relationships within it. A large language model (LLM), for example, learns patterns in how words and ideas tend to relate to one another. When you give it a prompt, it uses those learned patterns to generate a response.

One way to think about this is predictive text on your phone, but on a much larger scale. Predictive text suggests what might come next based on what you have already typed. A large language model works in a similar way, generating a response piece by piece based on your prompt and the patterns it learned during training.

This is why generative AI is not simply pulling a ready-made answer from a database. It builds a response in the moment, which means the same prompt can produce different results at different times.

Because generative AI can produce large amounts of content quickly and with very little effort, it has also contributed to what is sometimes called “AI slop.” This informal term describes high volumes of low-quality, repetitive, inaccurate, or poorly considered AI-generated content. Being able to generate something quickly does not necessarily mean that what is generated is useful, meaningful, or good.

Generative AI is no longer limited to working with text. Many AI systems can now understand and work across different types of information, including text, images, audio, and video. These systems are known as multimodal AI.

Working Across Different Types of Information

Multimodal AI can also work across different types of information at the same time. For example, you could upload a photo of a whiteboard full of notes and ask AI to turn the notes into a lesson summary, or provide a written story and ask it to create matching illustrations.

When working with generative AI, the instructions and context you provide can have a big impact on the kind of response you get in return. These tools don’t know your classroom, your students, or your teaching goals, they rely entirely on the instructions and context you provide. In the world of AI, these inputs are called prompts.

One way to think about prompting is like starting a conversation with a helpful colleague who doesn’t know your class yet. If you simply ask for “a lesson on fractions,” you’ll probably get something fairly general. If you add details about the grade level, learning goal, time available, or what students already know, the response can become much more useful.

Building a Clear Prompt

• Role: Give the AI a perspective or area of expertise to work from, such as an elementary reading specialist or a Grade 8 science teacher.
• Task: Clearly explain what you want it to do, such as create a quiz, adapt a reading passage, or draft a unit plan.
• Context: Add the details the AI needs to understand the situation, such as the grade level, learning objective, available time, or what students have already learned.
• Constraints: Include any boundaries or requirements, such as length, format, reading level, or tone.

Open Chatbots and Pre-Built Teacher Tools

Some education-focused AI tools that we will explore in Chapter 4: Daily Practice & Classroom Integration guide you through prompting by asking for information such as grade level, subject, learning goal, or type of activity. Instead of starting with a blank text box, you may choose from forms, menus, or a pre-built template which can make common tasks faster.

Open chatbots give you more flexibility to describe exactly what you need, ask follow-up questions, and refine the response over time.

In both cases, the same basic idea applies: the more useful context you provide, the more relevant the AI’s response is likely to be.

A Generative AI model is trained to respond to what you ask it to do. You give it a prompt, it generates a response and you decide what happens next. Agentic AI changes that interaction. Instead of guiding the AI through each individual step, you can give it a goal and allow the system to work through some or all of the steps needed to reach it.

An AI agent is a system designed to work toward a goal with some level of independence. It can decide what steps to take and use available tools to take actions on a user’s behalf.

What are the characteristics of Agentic AI?

• Works Towards a Goal: Rather than completing one request at a time, an AI agent can be given a larger goal, such as “help me organize parent-teacher conferences,” and determine the smaller tasks needed to work toward it.
• Plans and Selects Choices: An agent can decide what steps to take, what information it needs, and what to do next based on what happens along the way.
• Uses Tools: Agentic AI can be connected to other digital tools that allow it to do more than generate content. Depending on what it has access to, an agent might search the web, check a calendar, update a spreadsheet, or send a message.
• Completes Multiple Steps: An agent can carry information from one step to the next and use it to continue working toward the goal without needing a new prompt for every action.

How Much Can an Agent Do on Its Own?

Not every AI agent has the same level of independence. Some might complete a step and wait for you to review or approve it before moving forward. Others may be able to complete several steps on their own within the bounded domains it's been trained to operate within.

For example, an AI agent helping with parent-teacher conferences might check your available times, organize appointment requests, identify scheduling conflicts, and draft reminders. You might choose to review each step, or allow the agent to complete some routine actions automatically.

Moving from tools that answer simple prompts to autonomous systems that execute multi-step workflows is a significant leap. As AI continues to evolve, one essential truth remains the same: technology can help us execute tasks, but educators define purpose. Keeping your judgement, professional experience, and classroom context at the centre ensures that these tools remain supportive assistants, rather than decision makers.