“AI is going to take our jobs.” I hear that everywhere, and for a long time I aimed that worry at job titles myself. What moved it was running a software development company in Ho Chi Minh, Vietnam, and watching my own AI and my own people side by side every day. It isn’t job titles that get swapped for AI. It’s ways of working. Same title, two people, opposite fates.
In a meeting, one of them says: “What do you want to do? Just decide, and I’ll move on it.” The other, in the same kind of status update, says: “I’ve written out all the spec details. I’ve listed every open issue. The undecided pieces? Let’s leave them for now, draw the line here, and ship first.”
For a long time I filed the difference between these two people under “reliability.” Recently I found a more precise way to put it. The first one makes me build the answer from zero. The second thinks through ninety percent on their own, and brings me only the last ten. The decision.
This piece isn’t about which job category loses what percentage. It’s about what actually gets swapped out for AI, and in what order.
Our AI never asks “So what should I do?”
We run a large part of our day-to-day operations on AI (I wrote about how we let the team use Claude Code freely, and what that cost us). The first rule I ever wrote for that AI was one line long: “The CEO’s job is judgment only.”
The AI I raised on that rule organizes the information, compares the options, and brings me a recommendation with the reasoning attached. The only thing it withholds is the final sign-off. In other words: it thinks through ninety percent and asks for the ten. That reliable move, the one I described above. I trained it that way, so fine. That’s on me.
And here’s the thing I didn’t want to notice. “Just decide and I’ll move” sits one step behind what that AI already does. A report that asks me to build the answer from zero, versus a report that arrives with a recommendation. There’s nothing to compare. Yet for years, the only word I had for that gap was “reliability.”
This is the part I most want to say. What AI overtakes is less a job category than a way of working. We talk about it as a list of jobs. Translation goes, bookkeeping goes. Some jobs really will shrink as whole markets. But the order I watch things get swapped, inside my own company, runs along a different line. Work with no judgment in it gets swapped for AI first. Within the very same job title, you’ll find one way of working that gets replaced and another that doesn’t.
“Understanding someone” and “falling in line with them” are not the same thing
I don’t want to be misread here. This isn’t “stop checking with your boss.” We literally build our work on going out and asking customers why they bought, and why they didn’t. Going out to understand the other side is as basic as it gets. So what’s the difference between going to understand and going to fall in line?
The line is this: after you’ve understood the other person, is your own read of the facts, and your own judgment, still standing? The person who understands takes in your situation and changes their proposal. But they don’t change the facts. The person who falls in line reads your face and your position, and changes the conclusion itself.
There’s a test I run in my head. “If I had opened by saying the exact opposite, would this person have landed on the same conclusion?” If yes, that’s understanding. If I sense they’d have said the reverse — they’re falling in line with me. I don’t mind being disagreed with. What I mind is when there’s no judgment inside the person at all.
Once I was talking through a hire. I worried out loud that we might be about to load too much onto one particular candidate, and my dev lead did not say “You’re right, let’s pass.” He said: “I don’t actually know how they’d handle it — I haven’t spoken with them myself. Don’t decide on my take alone; interview them once. We could also split the workload and put a second person on as backup.” He took my concern on board and improved the proposal. But he didn’t let go of his own assessment that this was a person with a real sense of responsibility.
The opposite happened too, and it scared me. In a project retrospective, I let a negative impression slip first, and then, worried that everyone’s view afterward was bending toward mine, I went back to check. Maybe the person who falls in line is being manufactured by me, the one who says the strong thing first. I stand up here drawing the line, and I’m on the side that blurs it too.
And a report that falls in line has its conclusion already set, on my end; none of the person’s own judgment is in it. That’s the same side as “just decide and I’ll move” from the top of this piece.
The people above the AI are the ones updating reality
So if you just hold your own judgment, are you safe? It turns out that isn’t enough anymore either. Producing a recommendation with reasoning? AI does that too. When I feel “this person is above the AI,” it comes down, in the end, to one thing. They update reality.
Three real examples. When we tried to standardize on a single internal AI tool, one of my developers didn’t say “Sure, we can standardize.” He brought me the facts as they were: the tool people actually rely on for the heavy lifting is a different one; the recommended tool sometimes ignores instructions; and here’s what the team falls back on when they burn through the usage quota. I changed the plan because of that report. First-hand information neither I nor the AI had — and it changed the premise the decision rested on.
My QA team member didn’t come asking “Is this broken clock display a feature request?” She classified it herself as a logic bug, opened a ticket, requested the fix, tested it, and closed the loop all the way through to confirming it in production. For an issue she couldn’t reproduce, she didn’t stop at “I don’t know” — she asked the customer for a reproduction video. “So, boss, what should I do?” appears nowhere in any of that.
And there’s a report from a new grad on the team that I can’t forget. I’d asked them to do some data entry on a page. What came back was: “The entry is done. But when I look at the live page, the text is cut off partway, so the display side needs work.” Finishing the entry and finishing the page are two different things. That report moved one step — from the person who executes the instruction to the person who spots the gap with reality.
AI thinks. Our AI probably thinks faster and wider than I do. But AI can’t go look at reality. Go get the first-hand information on the ground, correct me or the AI when our premise is wrong, move people, and close the loop to a result. And beyond that, turn that experience into a system, so the same decision stops coming back to me next time. The value of a person is starting to shift from “thinking” to “taking on reality and responsibility.”
Isn’t this just convenient for the boss?
If you’ve read this far, some of you are thinking: “Asking employees to bring judgment is convenient for the boss. If you hand over neither authority nor recognition and still say ‘take ownership,’ that’s no different from exploiting people’s goodwill.”
Half of that, I think, is exactly right. If you demand sign-off from an employee who has no authority, that is exploitation. But sign-off isn’t what I’m asking for. I’m asking for the facts and the read. “This isn’t mine to decide. But I think option A is the way to go, and here’s why.” That’s enough — in fact, I think that’s what correct escalation looks like. I don’t want someone to take responsibility in my place. I want them to bring me enough facts and enough of a read that I can take that responsibility properly.
And if I’m going to ask for that, the company owes something in return. How far is someone allowed to decide? We haven’t drawn that line all the way yet either. Access to the information they’d need, protection that doesn’t punish an honest call gone wrong, recognition for the wider role: we’re short on all of it. If someone keeps asking “what do you want to do?” while we’ve handed over none of that, the cause isn’t the employee. It’s me.
Sometimes our AI’s recommendation looks more polished than the new grad’s report. But that’s not a fair fight. To the AI, I’ve handed over the decision criteria, the history, my preferred report format — patiently, over time. If I haven’t handed the same context and the same templates to the new grad, that’s a difference in onboarding, not in ability. We’ll debate carefully how to raise an AI, and then leave human onboarding at “watch and learn.” I suspect a lot of companies are like that. Mine included.
I decided to rewrite “the CEO’s job is judgment only” too
Writing this piece, I noticed the flaw in that very first line of my AI rule. “The CEO’s job is judgment only.” Written that way, every judgment climbs up to me. Which is, in the end, a company waiting on me to decide. The accurate version is: “The CEO’s job is only the judgments that can’t be delegated.” Funding, contracts, brand, personnel, promises to key customers. Escalate those to me. And for reversible calls that sit within someone’s authority, let each person decide, act, and report the result. Fixing the AI’s rule mattered less than writing down, explicitly, what I hand over. (The same way I wrote about drawing the line on what to hand to AI in the piece on sorting your data. Delegation to people also comes down to the same discipline: decide where the boundary sits, and revisit it monthly.)
One last test you can try in tomorrow’s meeting. When someone’s report ends with “what do you want to do?”, try asking back: “What do you want to do?” If a read comes back, that person came to understand. If silence comes back, then maybe the one who built a way of working with no judgment in it was us, on this side of the table. Before you rush to roll out AI (and there’s a trap where the chasing itself becomes the job), I think it’s worth checking.
FAQ
Whose jobs does AI actually take?
On the ground, it splits by way of working more than by job category. In my company, “just decide and I’ll move” reports had already been passed by an in-house AI that brings recommendations with reasoning attached. On the other hand, work that doesn’t stop at bringing judgment — work that goes out and corrects my or the AI’s premise with first-hand information from the field — is something we haven’t been able to swap out, at least for now. Not deciding “replaceable or not” by job title is the starting point for doing something about it.
How do I get my team to bring their own judgment?
Check what you’re handing over before you ask for it. The range they’re allowed to decide in usually gets stated; the rest often doesn’t — access to the information they’d need, protection that doesn’t punish an honest call gone wrong, or recognition for the wider role. In my own company, during the period when I said “take ownership” before handing any of that over, nobody moved. We now write down “mistakes are forgiven, hiding them is not” first, and then widen the range of decisions we delegate within someone’s authority.
Should I roll out AI first, or restructure the organization first?
I’ve found it works best to run them together. AI takes on the “organize, compare, recommend” layer, so rolling it out naturally surfaced the organizational question of “what do we leave to people?” The reverse also happened to me: by writing only “the CEO’s job is judgment only” and running the AI before spelling out the delegation range, I built a state where every judgment climbed up to me. If you’re unsure where to start in your own situation, it’s worth talking it through once.Running the rollout and the delegation boundary together is what our AI consulting work actually consists of.
Want to sort out “what to hand to AI, and what to leave with people” together?
Give me 15 minutes on how your company works today, and we’ll map out, right there on the call, which work you could hand to AI and which calls you want to keep with your people.
Book 15 minutes → or send a message
Shogo Harada原田 祥吾
CEO · Linnoedge Inc. · LinkedIn↗
Operating IT offshore development and overseas expansion support businesses across two bases: Tokyo and Vietnam. A leader who believes in “Systems over Spirit,” structuring cross-border businesses that often tend to be opaque. Committed to providing “reproducible quality” to organizations and clients rather than relying solely on individual skills.