AI agents are everywhere

If you follow AI news, you’ve heard the buzz around AI agents. You’ve heard they’re going to transform the way we work. You may have heard heated debates, in certain small circles at least, about the role of AI workflows vs agents.

So what actually is an AI agent? How do you use one in your day-to-day work? How does all that buzz apply to you?

In this first lesson, we will:

• Cover a practical and useful definition of AI agents

• Compare agents to other AI tools (that we should also all be using)

• Answer that agent vs workflow debate

• Share four common patterns of agents that you can apply to your work

AI agents, technically speaking

If you look up an expert definition of AI agents (that is, if you Google it, you might read:

“AI agents are software systems that use AI to pursue goals and complete tasks on behalf of users. They show reasoning, planning, and memory and have a level of autonomy to make decisions, learn, and adapt.” – Google

Or for a more precise definition:

“Systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks.” – Anthropic

These technical definitions talk about important aspects of agents: goals, tools, reasoning, and autonomy. But what makes them different from other AI tools that may do similar things?

The three modalities of AI interaction

There are three modalities of AI tool that we should all be using: Chatbots, copilots, and agents. They are all useful for slightly different purposes and manifest in different ways in our day-to-day work.

Chatbots

Chatting is the modality of interacting with AI that we’re most familiar with. You ask questions, and the AI responds. You direct it to do things, and the AI responds. Popular chatbots include ChatGPT, Claude, Gemini, and all those smart home devices we talk to.

Copilots

Copilots don’t live on their own, but in the context of another tool, and they help you take action in that tool. Grammarly is a popular example: It can help you write an email without leaving Gmail, for example. Cursor, another copilot experience, is a coding environment that helps developer productivity.

Agents

AI agents do work on their own, on our behalf, behind the scenes. Whereas chatbots and copilots require that you talk to the AI, an AI agent will wake up on its own, do work on its own, and achieve a real world task on its own. They can replace a meaningful amount of work without you needing to interact with it as it’s working.

A simple definition of AI agents

To capture the essence of the technical definition, and to differentiate agents from other common AI modalities, we propose this simple definition: An AI agent is an AI that works on its own, on your behalf, behind the scenes.

Now that you know what an AI agent is, and how it’s different from other modalities, let’s cover a more nuanced aspect of AI agents: How they differ from AI workflows.

The workflow vs agent debate

Think of a flowchart. Any task you accomplish in the real world, from making a sandwich to analyzing a thousand-row spreadsheet, can be represented as a flowchart: A first step, second step, maybe some logic or conditions, and so until the task is completed.

The difference between a workflow and an agent is who makes the flowchart and when.

In the case of a workflow, the flowchart is made ahead of time by the person architecting the workflow. 

For example, the architect would predefine rules that every time an email is received, the AI should check if it’s a cold sales email, and if so, auto respond with a polite declining the offer.

An agent, on the other hand, constructs its own flowchart on the fly based on its goal and the available tools.

For example, the architect would give the agent a goal of automatically responding to any cold emails, as well as tools to receive emails, decide if they’re cold sales emails or not, and auto-respond.

Now that you know the difference, let’s talk about why it doesn’t matter in practice!

1. Both AI workflows and AI agents both play the same fundamental role in your work. They both work on their own, on your behalf, behind the scenes, to accomplish a goal without you needing to do anything. Regardless of whether you predefined the flowchart up front, or if the AI defines it on the work, the same work is accomplished in the same way.

2. It’s a spectrum rather than a binary decision. In many workflows you want components to give AI the flexibility to determine what to do next. For many agents, you want to prespecify steps the agent should take to achieve aspects of the goal.

Since AI agents and workflows fill the same role, and it’s a spectrum, we’ll tend to just call all these AIs that do work on your behalf “AI agents,” since that gets the concept across more clearly.

Common patterns of AI agents

To make things more concrete, let’s talk about what AI agents do for us in our work. There are four common patterns that come up often when we’re building agents.

Pattern 1: Handling an internal event

This pattern is about processing information in a tool in your control.

1. When something happens in an internal tool

2. Ask AI to process the information with AI

3. Update another internal tool

Examples

Example: When you get a PDF invoice in your email, use AI to extract key data, and update a spreadsheet.

Example: When a new customer signs up for your service, look them up on LinkedIn, and add enrichment data to your CRM.

Pattern 2: Capture external updates

This pattern is about monitoring things in the outside world rather than your internal tools:

1. When something happens in the outside world

2. Ask AI to process the new data

3. Take action with the result

Examples of handling external events

Example: When a competitor posts a new YouTube video to their channel, analyze the content and send the summary to your team on Slack.

Example: When a new article is posted to an RSS feed, summarize the content and draft a social media post about it, then post it.

Pattern 3: Prepare for an event

Another common pattern is preparing for an event and gathering information or taking action before it happens (or following up afterwards).

1. When an event is upcoming

2. Look up relevant data

3. Process it with AI

4. Take appropriate action

Examples of preparing for events

Example: When a meeting is coming up, find LinkedIn profiles and past emails with guests, write a dossier, and send it to yourself over email.

Example: When a workshop you’re running is coming up, look up all the attending students and their project statuses, write them personalized emails, and send them reminders.

Pattern 4: Analyze on a schedule

This pattern of recurring analysis is particularly powerful because it’s relatively easy for AI to do, but very difficult for people to do efficiently. Chances are, you don’t do much of this in your work—yet!

1. On a recurring schedule (like every day/week/month)

2. Find information

3. Analyze it

4. Take appropriate action

Examples of recurring analysis

Example: Every day, look up all my team's customer call transcripts, synthesize insights, and send learnings to the team over Slack.

Example: Every month, look up popular LinkedIn posts form influencers in my space, analyze patterns, and email the analysis to yourself.

Example: Every quarter, look at competitors' pricing pages, check if they have meaningfully changed or not, and notify the team if they have.

Now you know

That’s it for lesson 1! Now you know:

• What AI agents are: AI that works on its own, on your behalf, behind the scenes

• The four major, useful patterns of agents in your business: Handling internal changes, monitoring external changes, preparing for events, and recurring analysis.

Next, we’ll start getting into the details of how you make these things, starting with the most important part of agents: the trigger.