There Are Two Ways to Use AI

There are two ways to use AI

One is to replace the laborious parts of your work with AI. The other is to rebuild the shape of the work itself, on the premise that AI is there from the start. Both are "using AI," but where you stand between them completely changes what comes out.

Ask people "Are you using AI?" and more and more answer "yes." Drafting meeting notes, writing first-pass emails, handing off part of the code. All of it is real use. But what's happening inside splits in two: are you replacing tasks while keeping your current way of working, or are you rebuilding the way of working itself?

And the feeling that "we brought in AI but nothing really changed" often comes from not separating these two — from doing only one of them without realizing it.

The two positions

Let me lay the two out a little more carefully.

Replace — hand part of the existing work to AI

Keep the existing process as it is, and replace the laborious tasks with AI. When you make a document, for example: tidy up the tables, draft the boilerplate text, clean up the meeting notes. The inputs and the deliverables stay the same as before; the work in between gets lighter. You don't change the shape — you just lower the load in the middle.

Rebuild — redesign the shape of the work around AI

Start from "AI is here," and redesign what you do and in what form. The input format, the granularity of records, who decides what — you rework all of it. Instead of tracing the old process, you bend the work itself into a shape AI can take.

The difference shows up clearly in three places: inputs, deliverables, and where judgment sits. Replace leaves these three in place and speeds up the middle. Rebuild touches the three themselves. So Replace is a sure first step you can take today, while Rebuild — because it involves remaking the work — has a bigger effect but takes more time.

What matters here is that Replace and Rebuild aren't opposites; they differ in reach. There's nothing to feel guilty about in choosing Replace. A sure, fast improvement is a valuable result in its own right.

Why "just replacing" tends to hit a ceiling

Still, the reason you hear "we don't really feel the change" is often that the work has unconsciously stayed within Replace.

The reason is simple: if you keep the shape of the process intact and only replace the tasks with AI, the bottleneck stays on the human-judgment side. Even if you hand only the clean-up to AI when making a document, the steps that decide "what goes in" and "is this the right direction" stay in human hands. However fast the in-between work gets, the overall speed still tops out there.

Widen the middle of a pipe, and if the faucet's diameter doesn't change, the amount of water coming out barely changes. It's not that Replace is meaningless — it's that the preserved shape of the process sets the ceiling on the effect. The real cause of "nothing changed" is often not AI's capability but where it was placed.

When you move to "Rebuild," a different job appears

So what happens when you move to Rebuild? One clear change: a different job stands up — remaking the work into a shape AI can take.

For the same document work, Replace speeds up the clean-up and the drafting, and stops there. Think it through as Rebuild, and the very order of "settle the goal, then start making" shifts. You change the form of the input so that even fragments from before the goal is fixed can be thrown in. You break down and redesign the granularity of what's recorded and where the human takes over the judgment. The role left to people then leans away from the hands-on part and toward sharpening the goal and making the final call.

This isn't limited to document work. In handling inquiries or in approval flows, the more you bend the work into a shape AI can take, the more the question shifts from "who moves their hands" to "where judgment sits."

Trace the history of work and this is a familiar move. Paper gives way to digital, steps get connected, mechanisms get rearranged — and on that same line is the stage of "rebuilding on the premise that AI is there." But if most past digitization traced the existing process to make it faster, rebuilding around AI moves the very location of who decides what. Where approval is needed, where the work stalls — that changes too. That's what's new. Rebuild, you could say, is the newest stage in this long stream.

Which to choose — purpose, cost, and who it lands on

Lined up this way, Rebuild might look like the strict upgrade, but it isn't. Which you choose is decided by purpose, cost, and who it lands on.

If the process you want to change is already well-ordered and all you want to boost is speed, Replace is often enough. Without paying the cost of remaking the work, a sure improvement is in your hands from today. On the other hand, if speeding up the middle has a visible ceiling, or the shape of the work itself has gone stale — when you feel that, the value of stepping into Rebuild appears.

Apart from purpose and cost, there's one more change worth watching. When you rebuild, the intermediate outputs stop appearing on their own. With Replace, the drafts and the in-between tables show up in your hands, so you can glance at them along the way and fix what's off. Once Rebuild makes the final output come out in one shot, where you check is no longer at hand unless you deliberately design it to be.

How this lands depends on where you stand. For someone who has judged by looking at the final output, the flow doesn't change much. But for someone who has worked by putting their hands in along the way, the impact is large. The very place to step in disappears, and the existing shape of their job can vanish with it. Handed a finished thing out of nowhere, you may want to fix it but can't see where to start.

So if you step into Rebuild, you end up redesigning not just the process but where people stand, together. Where to leave checkpoints, and which judgment to ask people to carry in place of the tasks that disappeared. Change only the shape without settling that, and efficiency goes up while the people on the ground are left hanging. The hard part of Rebuild is usually less the remaking of the process and more this re-placing of people.

So no line of superiority can be drawn between Replace and Rebuild. It comes down to how much reach your purpose actually needs. Bring Rebuild into a scene where a small speed gain is enough and it's overkill; take on a scene where you want to change the shape itself with Replace alone and it falls short. Where you place the tool — that difference alone splits what comes out, by a lot.

Which one is your workplace doing right now?

Once you're here, you can hold the two kinds up against your own workplace.

The "AI use" running in your hands right now — is it replacing tasks, or rebuilding the way of working? Make the question a little more concrete. Is what you want to speed up the task in front of you, or the shape of the work itself? Is the judgment left in human hands one worth sharpening, or one left over out of inertia? Past the point where you've sped up the middle, is there still headroom, or is the ceiling already in sight?

It's not about which is right. But being able to put into words which one you're doing makes it a notch clearer why "nothing changed" is happening, and which way to step next.

There are two ways to use AI: replace, or rebuild. Just being conscious of that one point makes the inside of the same "yes, we're using it" look quite different.