The job you're about to post may never need to exist

If you approve headcount, or if you are hoping someone approves headcount that includes you, this is the part of the AI conversation that actually touches your desk. It is not about a robot showing up and doing your job. It is about the job never getting written down in the first place.

Here is what happened to me. I was about to hire three staffers to gather data from all 254 county websites in Texas. Instead I set up a handful of AI helpers, named them after presidents, and had the data I needed in less than 24 hours. The job I was about to post never needed to exist.

That is the part people miss when they ask whether AI is going to take jobs. It doesn't take the job. It makes the job never exist in the first place, and that is what puts you on one side of the fence or the other: either you are the person making other people's roles unnecessary, or you are the person whose role never gets posted.

Why I keep describing this as an electric truck

It occurred to me while I was driving. I drive an electric truck, a Ford Lightning, and the right lane was closing while the guy in the left lane sped up to match me every time I did. So I tapped the accelerator, popped out in front of him, and was gone. It was easy. It was like nothing, because he was in a gas powered vehicle and I had that power sitting right there.

That is what AI does for me every day when I solve problems — but only because I am up to date with what is going on. I use the tools constantly. I don't work without one as my sidekick. When I need to solve a problem I have any number of tools available, I have my favorites right now, and I also keep an eye on the new things coming out. The pace is the real issue: there is so much out there that nobody can keep up with all of it, and each new tool has to be learned before you get anywhere near its full potential. So people who aren't keeping up aren't just slightly behind. The people who are learning this stuff have their foot on the same accelerator I had in that left lane, and they are going past. Some of them know many tools. Some have gotten extremely good at one tool and figured out how to run everything they do through it, and that is their superpower.

I have written before about the way we can build things now, which is the same idea from a different angle.

What actually happened when I handed the job to George Washington

I sat down in a morning meeting about a political website I am working on, with a political consultant. I told him it would be really helpful to go out and get data from all the different counties in the state of Texas. Texas has 254 counties, so I needed a way to gather data from all of them, across all these different websites and all these different file types — PDF, HTML, CSV, whatever it is. In plain terms: some counties publish a scanned or printed-style document, some publish it as a web page, some publish a spreadsheet-style file. Same information, wildly different packaging.

He said yes, we can find the funding, put it together, and find a staffer or two to go out and do that, gather the data and bring it together. And I said okay, but once that's done we also need someone who can sit there and normalize that data — meaning take all those mismatched formats and rewrite them into one consistent structure we can actually read and use — and put it into a file format we can understand.

Then I got home and thought, my friend just told me about Grok bots. Grok is an AI assistant, and a "bot" in this sense is an instance of it that you configure with a standing assignment and let run. Let me see if I can do that.

So I built a bot named George Washington and gave it the exact same charge I had given the person who was going to find me the staff and the funding. I literally said, "General, go to all 254 county websites in the state of Texas, look for this data, bring me this data."

Then I spawned a second bot — spawning just means creating another working instance — and this one was obviously named John Adams. I told it, "When General Washington is done, John, I need you to go through and normalize all the datasets as he's uploading them into this Google Drive."

Then I went back to George and gave it a really important command. I told it to spawn as many agents, as many bots, as it needed. An agent here is a worker the AI creates for itself to run a piece of the task in parallel, without me setting each one up by hand. Next thing you know I had Benjamin Franklin, James Monroe, Abraham Lincoln. I had all these presidents doing all this work for me. And it was really fun. It understood what it needed to do in order to satisfy the requirement.

The savings were not really payroll

I had done two things. I had removed the need to hire three staffers to go do this for me, and more importantly I had avoided the real cost of doing that, which isn't just three salaries: it's their insurance, their payroll and their taxes.

I want to be straight about the money, though. It wouldn't have been a lot of money overall. The most important thing to me was the time. I didn't have to train anyone. I didn't have to show them what to do. I didn't have to watch them get the data wrong and then go back and fix it. I had the ability to just punch the accelerator.

Once I spawned those agents and got them working, I had the information I needed and was back on the core project in under 24 hours. That was the best part, because it was going to take 24 hours just to come up with a job description to put out there, never mind going out and finding the people to do it. My speed to market — how fast I can go from "we need this" to "this is done and in use" — was so incredible that it really felt like tapping that accelerator and passing everybody else. That matters because as more people build programs and applications to solve problems, this lets them get to the root of the problem quicker, and to the root of solving it quicker, without slowing down.

Wasn't AI supposed to level the playing field?

There is a gap that wasn't supposed to be there. I believed the promise that AI would give everyone a level playing field, because it makes it easier for everyone to do all of the things. What I didn't take into account was that the people who are already really good were going to be so much better at doing things. The Texas project is my evidence for that: the difference wasn't budget and it wasn't headcount, it was knowing what to reach for.

So this is what I'm really looking at. It's not that AI is going to take the job. It's that AI is going to make the job not even exist, because AI lets you get so much done so quickly that the job won't even need to be there to be taken away in the first place. But to be on that side of it, you have to understand what's going on. If you don't, that's the lane I was talking about, closing on you while somebody else taps the accelerator.

What to do with this

• Before you approve the next hire for multi-step data gathering, ask how long it will take just to write and post the job description — and compare that to trying an agent-based approach first.
• Ask what the fully loaded cost of the role is, not the salary: insurance, payroll and taxes included. Then ask whether the real cost you are trying to avoid is money or time.
• Ask your team who has actually used these tools in the last month, on a real task, not who has read about them.
• Learn as much as you can about the AI tools available right now, not the ones you read about last year.
• Find a local friend group, local people you can talk to who are excited about this.
• Find one or two friends using it in ways you hadn't thought of, and push each other to go further.
• Get really good at at least one tool and figure out how to run everything you do through it, so that becomes your superpower.

Want this kind of thing weekly, in audio? That's the show.