Almost every week someone tells me they need an AI agent. I ask what they want it to do, and most of the time there's a pause. They've heard the word but haven't pinned down what it means.

"Agent" is having a moment the way "the cloud" did ten years ago. It floats around in meetings and strategy docs, and everyone nods along without really knowing what's underneath. So let's sort out what an agent actually is, and when you actually need one.

(Full disclosure, I have a couple of different agents, one of which manages the LA-AI social media accounts. Contrary to popular belief, it did not write this newsletter.)

Start with what you already know

You've used a chatbot. You type something, it responds, and then it waits for you again. It can talk, but it can't do anything on its own. No tools, no clock.

An AI agent is basically that same chatbot with two things bolted on: tools and a clock. Tools, so it can actually do things instead of just describing them. A clock, so it can run without you poking it every time.

That's really the whole idea. An agent is three simple pieces stacked together. Let's walk through them.

Ingredient one: a trigger

The trigger is what wakes the agent up, and it's the part most people skip right past. A chatbot only wakes up when you type. An agent can wake itself. Sometimes that's a timer, like every morning at 7. Sometimes it's an event, like a new email landing or someone filling out a form. It runs on a schedule or when something happens, instead of waiting for you to hit go.

What a lot of what people are calling "an agent" right now is just a process on a timer with better marketing. If all you've got is "every day at 6 PM, send me a report," that's not an agent yet. That's a reminder.

Ingredient two: a brain

The second ingredient is the AI part, the brain. When the trigger fires, the agent looks at the situation and decides what to do.

This is what separates an agent from ordinary automation. A plain automation follows a fixed script. Every time, no matter what, it does the same thing. Useful, but rigid. If something slightly weird happens, it has no idea what to do with it.

An agent makes a judgment call each time. Two emails come in, and it can treat them differently because it understands they're different. One's a calendar invite you can ignore. One's your biggest customer asking why their last invoice looks off. Those shouldn't get the same treatment. That flexibility is why agents can handle messier, less predictable work than a fixed script.

Ingredient three: tools

The third ingredient is tools. Tools are how the agent does things instead of just talking about them.

Connect an agent to your email and it can send a message. Connect it to your spreadsheet and it can update a row. Give it Slack, your calendar, your booking system, and it can post in a channel, add an event, or log a new customer. Whatever you've decided it's allowed to touch.

Without tools, even the smartest agent is just a chatbot with a clock.

This is where the risk lives too. Tools without clear limits is how you end up with an agent that emails the wrong customer, or overwrites the wrong sheet. So you don't just give it access, you set guardrails.

Put it together: the loop

Stack the three together and you get the loop. Trigger fires, brain decides, tools act, then it goes quiet until next time.

Say a timer goes off every morning at 7. That wakes the agent, which reads your inbox, picks out the messages that actually need a reply, and drafts responses for you to glance at over coffee. The timer was the trigger. Deciding which emails mattered was the brain. Reaching into your email to draft replies was the tools.

Swap one piece and you get a different agent. Trigger on a price change instead of a clock, and now you've got something that watches a product and pings you when it drops below a certain number. Trigger it on a new lead hitting your contact form, and you might have an agent that pulls up what it can find about them and drafts a first reply for you to check. Same three pieces, different job.

Once you start thinking this way, you see the pattern everywhere.

So do you actually need one?

Here's the part where I probably talk a few of you out of building anything. Most of the time, "I need an agent" really means "I have this one annoying thing I wish would handle itself." And often the answer isn't an agent at all.

Sometimes a simple automation does it. Sometimes a saved prompt you run yourself once a week does it. Sometimes the answer is "just make the spreadsheet better." Reaching straight for an agent can be like buying a bass boat to get across a parking lot puddle.

Agents earn their keep when three things are true at the same time. The task needs a judgment call, not the same rote action every time. It needs to happen on its own schedule, without you kicking it off. And it needs to touch your real tools, not just spit out text for you to copy.

If you've got all three, great, you probably do have a a good problem for an agent. If you don't, that's fine too. It just means the next step isn't hiring someone to build an "agent," it's getting specific about what actually needs doing. Either way, you're now a step ahead of most of the people using the word.

Know someone who’d enjoy this?

Don’t miss the next issue.