I am standing at the front of a conference room, counting on my fingers. Move the file. Build the chart. Send the email. Pull the data off the website. That is what an agent does, I tell the room. A chatbot gives an answer, and then somebody still has to go do it.
It is a good list. I have given it before.
The room is a product company with a few dozen people, and this is an AI lunch-and-learn. They sent me 16 questions ahead of time, but the one I am answering was not on the list. It came up live: if the company signs one enterprise AI contract, with every system plugged into it, does anyone still need to build agents?
It is not an idle question. Somebody in this room is going to approve that contract. If they buy it thinking the chatbot does the work, they will pay for a smarter conversation and still do every task by hand.
Heads nod. The question comes back anyway. Surely an enterprise license comes with company-wide agents already set up.
A colleague who has been researching this for the company takes a turn. Some of it can live at the company level, they say, where the shared connections to everyone’s systems sit. But most of the gain is personal, because everyone writes differently, researches differently, and needs different context to do their job.
I try an analogy. Agents are like spreadsheets. A company has shared spreadsheets that everyone works from, and every person also has their own, built for the way they work.
Closer. The room is almost there.
Then a hand goes up near the end of the table.
“Would it be a fair analogy to say a chatbot is more discussion, and an agent is more delegation?”
The room goes quiet.
I have been explaining this with lists.
They just did it in one sentence.
Discussion. Delegation.
“Yes,” I say. “That’s exactly it.”
That is the moment. Not because the line is clever, but because it is the line I should have opened with, and it came from the audience. I had been describing what an agent can do. They described what changes for the person using one. A chatbot is a colleague to talk things through with. An agent is a colleague to hand work to. Buying a better conversation partner does not take a single task off anyone’s desk.
One shared, one yours
Here is how the answer to the original question actually works.
Somebody still has to build the agents. No enterprise license knows how your company ships an order, answers a return, or reconciles a settlement report until someone teaches it. But once an agent is built, it can be reused across the company.
So think in two layers, the way you already think about spreadsheets. The company layer holds what everyone shares: the connections to your systems and the data they pull from. The personal layer holds what is yours: how you write, what you check, the context you need. You pull the shared pieces into your own agent, and when you build something the rest of the team needs, you share it back.
You may already be paying for it
At the time of writing, Anthropic includes its agent, Claude Code, in its paid Claude plans, and OpenAI includes its agent, Codex, in its paid ChatGPT plans. In both cases the agent shares the usage limits of the plan you already have. Free plans are a different story, and plans change, so check what yours includes. The pattern is the useful part: the agent is often a download away from the subscription you already pay for, and you only pay more if you use more.
Why the desktop matters
A web chat knows what you paste into it, plus whatever its memory feature has kept. A desktop agent can open the files on your computer: your notes, your past work, the instructions you wrote down last week about how you like things done. So it starts every session already knowing how you work, instead of waiting for you to explain it again.
Training wheels first
When someone asks me which platform to pick, my answer is: pick one and learn the basics. Ride with training wheels before you start running. Hand it one small, boring task this week, the kind you would give a new hire on their first day, and watch what it does.
The best line of that lunch did not come from the front of the room. It came from someone who had been listening all hour, and who saw what the rest of us were circling. Nobody in that room was short on opinions. They were short on hours.
Stop asking it questions. Start handing it work.
