The question that started all of this wasn't mine. It was my VP's, in a Tuesday planning meeting, and she asked it out loud to the whole room: if the agents could do every task this team does, who would we keep, and why?
Nobody answered. I went home and couldn't stop thinking about it, because I didn't have an answer for myself either.
If you searched something like "how to become irreplaceable by ai," you're probably somewhere near where I was that Tuesday. So here's what happened next, with real numbers where I have them. I run client operations at a freight brokerage in Columbus, nine people at the start of this year. In Q1 the company piloted AI agents on our quoting and tracking workflows, the two things that ate most of our day. Within a quarter the agents were handling both, mostly without us. By June the team was seven people. I'm still here, my title got a word added to it, and my pay went up $4,150 a year. Two colleagues I liked are gone. I want to be honest that both of those things are true at once, because every version of this story that only tells one half is selling something.
What follows is the six things I actually did between that Tuesday and now, in order. None of them required me to become a programmer. A couple of them were uncomfortable in ways I didn't expect.
One thing before the steps. When the pilot was announced I did what everyone does and turned to YouTube, three weeks of "will AI take your job" videos that contradicted each other completely, half of them panic and half of them cope. A short structured course on AI workflows cut through all of it in about a week, mostly because it made me actually use the tools instead of watching other people have opinions about them. YouTube is fine once you already know what you are looking for. It is a maze when you do not.
Also, the sixth step is the one that mattered most, and I'm keeping it for the end on purpose. It's the one nobody told me and the one I'd tell a friend first.
Step 1: Answer the keep-list question for yourself, on paper
I took my VP's question and made it personal. One page, two columns. Left column: every task I did in a week. Right column: honest answer to "could an agent do this within a year or two?"
The left column was longer than I expected and the honest answers were worse than I expected. Quoting, status updates, data entry between systems, most reporting: yes, an agent could, and within months it did. What survived the audit was a shorter list: talking a furious client down after a lost shipment, deciding which carrier to trust with a fragile load when the data said two of them were identical, being the person whose name was on the outcome when something went wrong.
I suspect most people never do this exercise because the first column is scary. Do it anyway. You cannot move your value until you know where it currently sits.
Step 2: Ask the people you serve what they'd actually miss
Here's the part I would have gotten wrong on my own. I assumed my value was speed and accuracy. So I asked four of our longest-standing clients a blunt question: if everything got automated tomorrow, what would you actually miss?
Not one of them said speed. One said "you answer the phone at 6am when a truck doesn't show." Another said, more or less, that she trusts me to tell her the truth when something's gone sideways instead of managing her. The talk that got me thinking about all this made the same point with a story about a sales team: when a company traced why its most loyal customers stayed, it wasn't price or product, it was how the reps made them feel. The company rebuilt the sales role around relationships and let the agents do the selling mechanics.
Your customers, internal or external, already know what your irreplaceable part is. It's probably not what your job description says. Ask them.
Step 3: Stop competing with the agents and start running them
This was the uncomfortable one. My instinct in Q1 was to quietly point out everything the agents got wrong, and for a few weeks I did, like a manager building a case against a rival. It felt productive. It wasn't. The agents improved every month and my list of gotchas got shorter, and I was positioning myself as the person who resists the tools instead of the person who runs them.
The switch was volunteering to own the agent workflows: reviewing their outputs, catching the genuinely weird failures, deciding when a quote needed a human pass before it went out. That's the job now, honestly. The agents do the doing, and I'm accountable for what ships. Accountability turned out to be the thing that doesn't automate, because when a $40,000 shipment goes wrong, the client doesn't want an explanation from a model. They want a person whose job is on the line.
Step 4: Put in actual weekly learning hours, like freelancers do
A stat from that talk stuck with me, and I'll pass it along with the caveat that I'm quoting a talk, not my own research: freelancers apparently average around four hours a week learning new skills. Employees average close to none. That gap made immediate sense of my own team, because the two colleagues we lost were, and I say this with real affection, the two who most confidently said the agent thing would blow over.
I blocked two hours, twice a week, before work. Not inspirational-video hours. Hands-on hours, building the same workflows the pilot used, breaking them, learning where they fail. I used a short course on Mindwand for the structured part, 15-minute lessons that fit before my first call, though Skool communities and Coursera cover similar ground and the format matters more than the brand. Pick one. Finish it.
Step 5: Make your judgment visible before someone asks for it
Invisible judgment gets cut with the task list. If the valuable thing you do is decide, vouch, and catch problems early, and none of that is written down anywhere, then on a spreadsheet you look identical to the tasks the agent just absorbed.
So I started documenting decisions, lightly. When I overrode an agent's carrier pick, I wrote two sentences on why. When the 6am phone call saved an account, the account manager heard about it. This felt like self-promotion at first and I hated it. But when the June restructure came, my manager didn't have to guess what I did that the agents couldn't. There was a trail.
Step 6: Reposition before the org decides for you
Here's the one I saved, and it's the difference between my outcome and my two colleagues'.
Every restructure has a moment where roles get redrawn, and you are either holding a proposal at that moment or you are subject to someone else's. About six weeks before decisions got made, I wrote a one-page pitch for what my role should become: owner of the agent workflows, escalation point for the relationships that keep our biggest accounts, trainer for whoever touches the tools next. I gave it to my manager before anyone asked.
I don't know for certain that the page is why I stayed. Maybe I'd have stayed anyway. But my colleagues waited to be told what their jobs would become, and the org answered honestly: nothing. Jobs are fixed things and companies will protect them exactly as long as they make sense on a spreadsheet, which is not long right now. What doesn't show up on the spreadsheet is a person who walks in with the redesigned version of their own role. From what I can tell, almost nobody does this. That's precisely why it works.
The talk I keep mentioning made an argument I think about a lot: the honest move isn't asking whether AI will keep climbing, because it will and that part is not up to you. The part that's up to you is how deliberately you move while it climbs. If you want the fuller argument about whether your specific job is exposed, I wrote up my read of the actual displacement data in another post, and the structured-learning piece of step 4 is the same Mindwand, Skool, or Coursera decision I mentioned above.
None of this took genius. It took about five months, two early mornings a week, four slightly awkward client conversations, and one page of writing my manager didn't ask for. The two hardest parts were emotional, not technical: admitting how much of my old job was automatable, and watching people I liked not make the same moves in time. I hope some of it lands.