The Intern

9 minute read

I hired an intern a few months ago.

Not a person. A system. A set of AI sessions that run around the clock, doing real work alongside me whether I'm at my desk or asleep. They build things. They research things. They manage my calendar and my household and my side projects and the tool I'm building and the newsletter you're reading right now. I didn't give them a manual. I gave them goals. And every day they show up, figure out how to get it done, and report back.

I know that sounds like I'm being cute with the metaphor. I'm not. What I just described is the single most important shift happening in AI right now, and the word for it is everywhere while the explanation is nowhere. The industry picked up "agent" and kept walking, and the rest of us got left standing there nodding like we understood.

So let me actually explain it. Because once you see it, you can't unsee it, and the way you work with these tools will change the moment you do.

The Word Nobody Defined

Agent.

You've heard it. You've read it. It's in every headline, every product launch, every breathless LinkedIn post about the future of work. AI agents are here. Agentic workflows are the future. The age of agents has arrived.

And if I asked you to define it, right now, in one sentence, most of you would hesitate. Not because you're not smart. Because the people using the word the loudest never stopped to say what it means. It's one of those terms that got popular before it got explained.

That's what I'm here for. So here it is, as plainly as I can say it.

An agent is an AI that doesn't just answer. It acts.

You give it a goal. It figures out the steps. It picks up tools, uses them, looks at what happened, decides what to do next, and keeps going until the goal is done. Nobody scripts the steps in advance. Nobody tells it which tool to use or in what order. The goal comes from you. The plan comes from it.

That's it. That's the whole idea. And it changes everything about how you work with AI once you understand what it actually means.

The Checklist and the Goal

Let me show you the difference, because the difference is the whole point.

Recently, I was dealing with a situation at home. Something had happened in my driveway, caught on camera, and I needed a license plate from the footage. So I handed the problem to Claude. Not a detailed prompt. Not step-by-step instructions. Just the goal: can you get a license plate from these videos?

And then I watched.

Claude probed the video files to figure out what it was working with. It extracted frames. It looked at them. It noticed a truck. It cropped the image. It enhanced it. It realized the angle was bad, so it went looking for a better frame from a different camera. It found one, re-cropped, enhanced again, chased details I hadn't thought to ask about. It abandoned dead ends without being told to. It tried approaches I wouldn't have thought of. Dozens of steps. Decisions at every turn. I gave it a goal and it built an entire investigation around it.

Now compare that to what most people think of when they think of AI.

You open ChatGPT. You type a question. You get an answer. One shot. You send text in, you get text out. That's not an agent. That's a very smart search engine. It responded. It didn't act.

The license plate wasn't a response. It was a project. Claude had tools — it could extract frames, process images, crop, enhance, analyze. It had a loop — after every step, it looked at the result and decided what to do next. And it had a goal — the license plate — that it kept working toward through dead ends and pivots and judgment calls.

The formula people in the field use is simple: agent equals a model, plus tools, plus a loop, plus a goal. That's the recipe. Each ingredient matters. Take away the tools and it can only talk. Take away the loop and it can only respond once. Take away the goal and it's just running in circles. Put them all together and you get something that acts. Something that works.

Something that looks, honestly, a lot like a sharp new hire on their first week.

The Intern

I keep coming back to this metaphor because I think it's the most honest one available.

When you hire an intern — a real one, a person — the first week is checklists. You write everything down. Step one, do this. Step two, do this. Step three, do this. You don't trust them yet, not because they can't think, but because they don't have context. They don't know how you work, what matters, where the landmines are. So you script it. You micromanage. You hand them a checklist and you check every box behind them.

That's prompting. That's what most people are doing with AI right now. "Write me an email that does X, in this tone, with these three points, formatted like this." Every instruction laid out. Every step spelled out. A detailed script for a system you don't fully trust yet.

And for a while, that's the right approach. That's how you learn. That's how the relationship builds.

But a good intern doesn't stay on checklists forever. A few weeks in, something shifts. They start to get it. They understand the context. They know what you care about and what you don't. They can anticipate. And one morning you stop writing the checklist and you just say: handle the vendor meeting prep. And they do. They pull the contract, check last year's notes, flag the pricing change, draft the agenda. Not because you told them each step. Because you told them the goal and they had enough context and enough capability to figure the rest out.

That's the shift from prompting to agency. From scripts to goals. From micromanaging a tool to managing a thinker.

But here's the difference between a human intern and an AI one. A sharp new hire will start making those judgment calls on their own. They'll just do it, because that's what people do. An AI won't. It waits for you to open the door. It waits for you to say "you have permission to think here." And most people never say it. Not because the AI isn't ready. Because they aren't. We can talk about agentic capability all day, but none of it matters until you're comfortable enough, and brave enough, to stop scripting and start trusting. The technology was ready to be agentic before we were ready to let it. And for a lot of people, it's not even a trust problem. It's a comprehension problem. They didn't hold back because they weighed the risk and decided to be cautious. They held back because they didn't know this mode existed. You can't hand over the reins to something you didn't know could run.

The Death of the Script

I want to say something that might ruffle some feathers, but I think it's true and I think it needs to be said.

Prompt engineering, the way most people learned it, is becoming the checklist you outgrow.

Not worthless. Never worthless. The discipline of being clear, being specific, knowing what you want and how to say it — that's permanent. That's the foundation. I wrote a whole post about how words have weight and every one of them matters. I stand by that.

But the skill of writing a detailed, multi-step prompt that scripts exactly what the AI should do? That skill has a shelf life. Because the AI doesn't need the script anymore. The models have gotten good enough to take a goal and run with it. They have tools. They have loops. They have judgment. They can decide what to do next without you laying out every step.

The skill that's replacing it isn't less important. It's harder. It's knowing what you actually want. It's defining the outcome without over-defining the path. It's setting the goal clearly enough that a smart system can work toward it, and loosely enough that it can use its own judgment along the way.

That's not prompt engineering. That's management.

And here's the part I find fascinating: the people who are going to be best at this aren't necessarily the most technical people in the room. They're the people who know how to delegate. Who know how to brief someone. Who know what "done" looks like and can describe it without scripting every step to get there. The skill the AI world spent two years telling you was the future — prompt engineering — turns out to be the training wheels for the actual skill, which is running a team of systems that think for themselves.

The Spectrum

This is the part where I pump the brakes a little, because I don't want to oversimplify this.

It's not a clean line between "prompt" and "agent." It's a spectrum, and most of the useful work happening right now lives somewhere in the middle.

On one end, you have pure prompting. One question, one answer, no tools, no loop. That's the chatbot experience. It's useful. It's not going anywhere. Sometimes you just need a quick answer and a single shot is exactly the right tool for the job.

On the other end, you have full agency. An open-ended goal, a system with tools and judgment and memory, making its own decisions in real time. The license plate investigation. A research agent that chews through a significant backlog while I sleep. Or this one, which still surprises me: when someone reports a bug in one of my products, the system automatically logs a ticket, assigns it to the right agent, and that agent picks it up, diagnoses the problem, fixes it, writes the release notes, emails the people who need to know, and deploys. No human in the loop. A bug report goes in one end and a fix comes out the other. Think about how many people and how many steps that process takes in a traditional company, and then sit with the fact that an agent does it while I'm doing something else entirely.

In between, there's a whole world of workflows — systems with predefined stages where AI fills in the work at each step. Think of it like a relay race with a set number of legs, but each runner decides how to run their stretch. The stages are scripted. The execution within them is agentic. Most production systems work this way, and it's a good way to work, because it gives you the reliability of a structure with the flexibility of intelligence inside each step.

The point isn't to live at the extreme end of the spectrum. The point is to know the spectrum exists. Because right now, most people are sitting at the chatbot end, writing detailed prompts, getting single responses, and they don't realize there's an entire other mode available to them where the AI does the planning, not just the answering.

What Changes When You See It

Once you understand the difference between a response and an action, between a script and a goal, something shifts in how you sit down with these tools.

You stop writing longer prompts and start writing clearer goals. You stop thinking about what you want the AI to say and start thinking about what you want it to do. You stop scripting and start delegating. And the first time you hand an AI a genuine goal and watch it build its own plan and execute it, making decisions you didn't anticipate, using tools in combinations you wouldn't have thought of, recovering from dead ends without being told — something clicks. It feels different. It feels like the first time you trusted a colleague to own something instead of just follow your instructions.

That's not a small shift. That's the shift. And it's happening right now, to everyone who works with these tools seriously, whether they have language for it or not.

I've been running my life this way for months. Multiple sessions running at once, each with a goal, each figuring out how to get it done. I set direction in the morning. I check in during the day. I get reports at night. I'm not writing scripts. I'm managing interns. Some of them are brilliant. Some of them need more guidance than others. All of them are getting better. And the thing that's getting better fastest isn't the AI. It's me, learning how to let go of the checklist and trust the goal.

Why This Matters for You

If you're reading this and you've only ever used AI as a chatbot — typed a question, gotten an answer, typed another question — I want you to know something. You're using a race car to drive to the mailbox.

That's not an insult. Most people start there. I started there. We all start there, because that's the mode that's obvious, and nobody explains that there's another one.

But there is. And the gap between where most people are and where these tools can take them is the gap between a checklist and a goal. Between typing a prompt and handing over a project. Between talking to your AI and working with it.

You don't need to be technical to make this shift. You need to be clear about what you want. You need to be willing to let go of the step-by-step. And you need to try it, just once, the way I tried it the night I said "can you get a license plate from these videos" and watched something I didn't expect happen on my screen.

The intern is ready. The question is whether you're ready to stop handing it a checklist and start handing it a goal.

The Part I Didn't Expect

There's a side effect to this that caught me off guard.

When you shift from scripting to managing, the thing that changes most isn't the AI. It's you. You start thinking differently. You stop breaking problems into AI-sized pieces and start thinking about what you actually need accomplished. You stop asking "what should I tell it to do" and start asking "what do I need done." That sounds like the same question. It isn't. The first one keeps you in the weeds. The second one makes you the manager.

I spent twenty years translating between engineers and executives, and I never imagined that skill would become the most important thing I do with a computer. But here I am. Briefing systems. Setting goals. Checking in at the right moments. Knowing when to let it run and when to step in. Managing.

The AI got promoted. It's not answering questions anymore. It's doing work. And that means you got promoted too, whether you realized it or not. You're not the person at the keyboard anymore. You're the person who knows what needs to happen and trusts something else to make it happen.

That's the intern story. Not a tool that got smarter. A relationship that grew up. From checklists to goals. From scripts to trust. From prompting to managing.

And honestly? It's the most interesting thing that's ever happened to the way I work.

Dacia writes about AI for real people at Speak Human. If you're trying to figure out how to actually use these tools in your everyday life, you're in the right place.