While you're chasing AI, AI doesn't get deployed

Hi, I’m Shogo Harada, CEO of Linnoedge, writing from Ho Chi Minh City.

I’ll give you the conclusion first. While you’re chasing AI, AI doesn’t get deployed.

There may well be a comparison deck open on your screen right now. A new model shipped. A company in your industry apparently rolled something out. Someone upstairs has asked what the plan is.

So you read, you compare, you run a trial, and that goes on for months — and the way the work gets done hasn’t moved much.

That’s the stuck point I keep hearing about from the people who run IT inside a company.

“AI fatigue” is a phrase I hear more and more.

It isn’t only the people around me who are stuck between using AI and actually deploying it. That’s the separation worth making before the numbers land. “We use AI” and “AI is deployed into how the work runs” are two different sentences, and the distance between them is what this article is about.

Stanford’s 2025 AI Index Report, drawing on McKinsey’s global survey, found that 78% of respondents said their organizations were using AI in 2024, up from 55% the year before.

Where financial impact gets reported at all, it’s thin. Among respondents whose organizations use AI in service operations, 49% report cost savings — and most of those put the saving at less than 10%. In marketing and sales, 71% report revenue gains, and the most common level of increase is under 5%.

Plenty of people are using it. The impact they report is still thin. That distance — between using it and changing how the work runs — is what I mean when I say AI doesn’t get deployed.

Let me draw a clear line here. The Index measures two things: how many companies use AI, and how big the reported financial impact is. It never says anyone is tired. It never says the reason is that they’re chasing.

What I can borrow from the data stops at one fact: a lot of companies are using AI without much changing. The explanation from here on is mine, not Stanford’s.

What I see is that the research time quietly takes over. The decision keeps not landing, and the next comparison starts.

One definition before I go further. When I say “chasing” in this article, I mean staying on top of the newest tools, models, and prompts. I’m not telling you to stop reading the news.

I use AI every single day, and I keep my own rulebook: plain markdown files where I’ve written down how I work, which I rewrite most weeks. Using it hard is the baseline, not the debate. My point is narrower: as long as that’s all you do, the work inside your company stays exactly where it is.

So I won’t tell you which tool is strongest or which model is fastest. That will have changed by next month, so writing it down is a waste of your time and mine. I want to talk about what you chase instead.

The problem isn’t the amount of chasing. It’s the object of it.

My guess at why the chase is draining: what you chase has an expiry date

You already know this one. An engineer spends years getting good at a language, the language falls out of favor, and most of that effort turns into sunk cost. If you sit on the IT side, you’ve lived it.

How to drive a specific tool. The newest spec of a language. Domain knowledge that only travels inside one industry.

These skills share a trait: they’re expensive to acquire and they go stale every time the ground shifts. That quietly wears you down.

The same thing is happening now with prompts. People collect them, memorize the good ones, drop them in a shared folder. Then the model takes one step forward. Most of those prompts don’t stop working so much as stop being needed.

That’s where the fatigue comes from, I think. Work that leaves nothing behind is draining on its own. It’s simply hard. There was a stretch where I half-knew that what I’d learned this week would be worthless next month.

And then it becomes “I can’t decide yet, I need to look into it a bit more.” Nothing accumulates, so you never build the footing you’d need to decide.

“Stop chasing and you get left behind” is correct

Some of you are thinking: hold on. In AI the assumptions flip week to week. Stop chasing and you’re behind the moment you stop.

I agree. I feel it too. In this field, what was obvious yesterday is wrong today, routinely.

But the word “chasing” is carrying two different things at once, and while they stay mixed, the argument goes nowhere.

One is chasing tool usage, prompt patterns, new features.

The other is chasing the structure of how AI changes the work itself.

The first resets to zero every time the ground shifts. The second compounds every time the ground shifts. Same effort, opposite direction.

Same word, opposite destination. I’m betting on the second one.

Precisely because I’m afraid of being left behind, that’s the side I’d rather run on.

So: change what you chase, from the things that expire to the things that don’t. That’s the whole article.

What exactly is the knowledge that doesn’t expire?

What I’m betting on is knowledge that carries over when the tools change. Two habits, specifically. Whether it survives the next few years is something I can’t promise you — I’m testing it on myself right now.

It’s the same bet I made when I asked why you’d still keep an offshore team once AI can write the code: the convenience is available to everyone, and the structure around it isn’t.

The first is reading the structure of the work. What is this process actually made of, and which parts of it don’t need a human? The second is what makes the first one usable: modeling — writing the tacit way you do things into something you can hand over.

Neither of those has a tool name in it. What they operate on is the shape of the work itself, and that doesn’t get replaced when the tool underneath it does — which is the whole difference from a prompt library, where the object of the skill leaves with the model.

What I’m calling durable knowledge today might still turn out to be the stuff that just happens not to have expired yet.

The reason I spend my hours there is that it keeps me out of the noise around any particular tool or prompt. Honestly, that’s all I want out of it.

Back to that rulebook. I have about ten of those files now, and if I switch platforms they might all become useless overnight. That scares me too.

But the hours I spent writing them — asking where my own work jams up, and what I’d have to break apart before I could hand it to anyone — those don’t switch off. What survives isn’t the file. It’s the habit.

And once you have that habit, you can decide which gap to close without reading the comparison deck to the last page. That’s the footing I mentioned earlier.

You build the habit on the floor

Reading structure isn’t something I learned in a classroom — I picked it up on the floor. The entrance is this: pick one gap in your workflow and close it, no comparison round. That’s it. Deciding which step doesn’t need a human is the practice.

You know better than I do that a core system never swallows the work whole — the gaps that just came to mind while you read that are the list.
The hard part is choosing which one to close first. Whether AI can even work in your environment yet is a separate question, and I’ve written about that in is your team ready for AI-native development.

The first gap I closed was project management. With several engineers running in parallel, you lose track of what finished today, what got added, and what slipped. Keeping ticket states honest, in other words. It’s unglamorous, and when it breaks, the whole project breaks with it.

It used to take me over an hour, every time. Handing it to AI, it now takes ten minutes of review.

Not zero. A human still looks at the end of it. But an hour became ten minutes. A quiet change.

I’m the only approver here, so closing that one gap took no model bake-off.

Choosing a platform for the whole company is a different animal. Security, data leaving your walls, licensing, sign-off — that job genuinely requires comparing vendors.

Closing gaps is not a substitute for it. It’s a different job.

Handing out a prompt library does work as an opener. Nothing starts until people have actually touched the thing.

The trap is stopping there, because the next year tends to start from the same place.
What you leave behind after the handout is what carries. The habit of breaking work apart and modeling it travels with you when you switch tools.
I’m still mid-way through breaking down my own job, sharing the parts that are written down, and rewriting about half of it the following month.

If sign-off is what stops you, one thing is worth noticing. If your company has already approved any AI at all, even a chat assistant, that tool has already cleared the approval bar. Closing a gap inside it needs no new procurement decision, which is why it can start before the platform question is settled.

Take the AI you already have. Close one gap in front of you. Park the company-wide rollout. One gap.

Once a gap is closed, it becomes a place you no longer have to chase — a new model can land next month and that work keeps running anyway. Do that a few times and the number of places you don’t have to chase quietly grows. While the comparison is still running, that number stays where it is.

Which is where we came in. Change what you chase, close a gap without a bake-off. That’s how AI started getting into my company, slowly.

Tomorrow’s decision isn’t “which tool”

If you’re staring at a comparison deck right now and can’t land the decision, swap one question.

Not “which tool do we bring in,” but “which gap do we close first.”

The answer to that one isn’t out there in the industry news. The gaps on your floor don’t move on their own when a model ships, so you stop having to go looking for answers. Since I switched to that question, I spend a lot less time watching the feed.

Less time — not the right answer. I haven’t arrived anywhere.

Chase the things that expire, and time passes with nothing left in your hands. That was me, in that stretch I mentioned. Change one thing about what you chase and see what happens. I’m in the middle of that experiment myself.

FAQ

If I stop following the news, won’t I become the IT lead who knows nothing?

This isn’t about stopping. It’s about what you follow. Drop the goal of catching every new feature, and read the same news with one question in mind: which gap in our work does this change close? Same feed, different residue. And once you’ve actually closed one gap, that’s one thing you can talk about concretely, instead of in general terms.

So building an internal prompt library is a waste of time?

Collecting them works, I think. The catch is that what you collect has a shelf life. “This phrasing gets better output” is the perishable kind. “This process breaks apart like so, and that’s the part we can hand over” survives a tool switch. If you’re going to write something down either way, write down the second kind — that one you get to keep.

How do I pick the first gap?

Pick for clear edges over big impact. Somewhere the input and output are defined, and where a failure doesn’t stop the business. Work where a person is copying data by hand, or running the same check every time, is usually easy to spot. Start with something big and you tend to stall in requirements — and then the time flows back into research.

We can’t isolate our own gaps from the inside. Can we talk it through with you?

Yes. The people inside your company already know where the gaps are, more precisely than any outsider will. What someone from outside can do is sit with you while you pick the first one — that call is heavy to make alone. We look at how the work actually flows, and we pick one, on the call.

Let’s close one gap you never have to chase again

Give me 30 minutes. We’ll pick one gap in your workflow and design how to close it, with no vendor bake-off. All we talk about is choosing the first gap. Think of it as a sounding board for adding one place you no longer have to chase.

Book 30 minutes → Message us instead →

Shogo Harada原田祥吾

CEO · Linnoedge Inc. · LinkedIn↗