AI show-offs and secret users: the two schools of AI at work

Good evening. The Owl here.

Lately I’ve been hearing that two schools of thought have formed among people at the companies leading the AI race, and among those who work around them.

The first is made up of people who actively advertise how much they use AI.

They push AI adoption inside the company, run study sessions, and present results at executive meetings. They see anyone who doesn’t use AI as behind the times, and sometimes push AI into the work even when it doesn’t suit the task or the realities on the ground. If adoption rates and prompt counts go up, they win. Even when there are no results, the first move is to declare the company “AI-first.”

Let’s call this the Nanto Seiken school, the one that shows off its moves with a flourish.

The second is made up of people who use AI very skillfully but never talk about it at work.

If people found out they had automated their job, their position might become a target for cuts. Even if it didn’t, they might just get handed more tedious work to fill the free time. They could be seen as people who simply use AI to slack off, with no guarantee of credit. And in quite a few cases, because the company-approved AI is hard to use, they quietly use the latest, officially banned AI through personal accounts or their own devices.

This is the Hokuto Shinken school, passed down quietly in the shadows of the age.

(Both names come from Fist of the North Star, the classic Japanese martial-arts manga: Nanto Seiken, the “South Dipper” style, is a showy family of 108 branches, while Hokuto Shinken, the “North Dipper” style, is a secret assassin’s art passed to one successor per generation.)

Of course, this isn’t an official classification. It’s just a metaphor.

But when I looked into it, I found that this joke of a split comes surprisingly close to what’s happening in real workplaces.

---

Nanto Seiken: a world where using AI earns you credit

Let’s start with the Nanto side.

It’s no longer unusual for executives to publicly champion AI and ask their employees to use it too.

In 2025, Shopify CEO Tobi Lütke made using AI effectively a baseline expectation for every employee. He also asked teams to show why AI can’t do the job before they ask for more headcount or resources. [1]

Amazon CEO Andy Jassy also stated plainly, in a message to employees, that the spread of generative AI and agents will change how work gets done and that the company’s corporate workforce will shrink over the next few years. At the same time, he urged employees to learn about AI, take training, and experiment wherever they can. [2]

There’s real logic to this thinking.

A new tool won’t spread through an organization unless someone waves the flag. If only the people who use AI get faster on their own, while business processes and product design stay the same, the company as a whole doesn’t become much more competitive. Leaders do need to clearly reward AI use and to provide the training and tools to make it happen.

A 2024 survey by Microsoft and LinkedIn also reported that AI power users were more likely to hear from leadership about the importance of generative AI and to receive role-specific training. [3]

The problem starts here.

Once using AI stops being a means and starts becoming proof of loyalty, an organization can go wrong very easily.

Nanto runs amok: usage rates masquerade as results

At companies that push AI too hard, the first thing that gets judged is whether you used AI.

So employees wedge AI into work it isn’t suited for. A piece of text that would take ten minutes to write normally takes twenty: explain the task to the AI, fix the output, check the facts. But because you can still report “I used AI,” the company’s usage rate goes up.

The executive dashboard shows adoption numbers, user counts and prompt counts all climbing. Customer satisfaction, quality and profit, meanwhile, barely move.

A 2026 survey by the Glean Work AI Institute shows this gap quite vividly. While 87% of digital workers use AI at work and 75% feel more productive, only 13% said AI had significantly improved their organization’s performance. The survey also found that workers spend an average of 6.4 hours a week checking, fixing and re-entering AI output to make it usable. [4]

Individuals feel faster, yet the organization hasn’t become much stronger.

What fills that gap is invisible cleanup work done by humans.

On top of that, BetterUp Labs and the Stanford Social Media Lab use the word “workslop” for AI-generated work that looks polished but has little substance, pushing the job of fixing and re-researching onto whoever receives it. In a survey of 1,150 US desk workers, 40% said they had received workslop in the past month, and that each instance took an average of two hours to deal with. [5]

That is what makes Nanto Seiken so dangerous.

Flashy AI documents can be produced in an instant. But the work of checking their assumptions, verifying the numbers and turning them into something you can show a customer has moved to someone else. The author looks more productive, and only the recipient’s workload grows.

Pushing more AI into a company doesn’t necessarily make it stronger. If the work of showing off AI use keeps growing, AI may actually make the company busier.

And recently, exactly this problem played out inside Amazon.

Amazon had an internal leaderboard called “KiroRank” that ranked employees by how many tokens they consumed using Kiro, its AI tool for developers. It was an unofficial system built by a group of employees, but some staff started “tokenmaxxing,” having the AI run tasks of little real need just to climb the rankings. Usage went up, business value didn’t, and only the compute bill grew.

Amazon shut the leaderboard down, and executives told employees not to use AI for its own sake but to use it to solve customer and business problems. The company said the ranking was never meant to be used to judge employees’ AI use in performance reviews. Even so, put a visible number in front of people and they will optimize for it. Not even Amazon, a company that sells AI and runs AI infrastructure, could avoid this trap. [6]

---

Hokuto Shinken: the most skilled users go quiet

So why does the Hokuto side hide?

One reason is that there’s no guarantee the gains from working more efficiently will come back to you.

Say you use AI to shrink an eight-hour job down to two hours. If those six hours became your free time, sharing the method would be worth it. If it raised your evaluation or pay, you’d want to teach your colleagues too.

In reality, though, it might be used as grounds for job cuts. Or you might be told, “Since you have time to spare, you can handle this too.” If the company captures every bit of the efficiency gain, the most rational move for the individual is to stay quiet about finishing in two hours and pretend it took eight.

Another reason is that using AI still carries a social stigma.

A study published in PNAS in 2025, based on four online experiments with 4,400 participants, found that employees who use AI tend to be seen as lazier, less competent and less diligent than those who don’t. In hiring scenarios, too, candidates who disclosed using AI tended to be at a disadvantage. [7]

This shows that the Hokuto side’s caution is not just paranoia.

Even when the quality of the work is the same, the moment people learn “AI made this,” they discount the person’s effort and ability too. A company may say it values people who can use AI, but the managers and colleagues on the ground don’t necessarily share that view.

So the people who use AI best keep it backstage, like a stagehand dressed in black.

Using it even when it’s banned: “shadow AI”

Hokuto Shinken has a somewhat more dangerous form, too.

The AI the company provides runs an outdated model. Requests for access take ages. It can’t connect to internal data. Competing services are banned. Yet with a personal subscription to ChatGPT, Gemini or Claude, you can use the latest, most capable models right now.

The result is that employees drift toward personal accounts and personal devices.

In the 2024 Microsoft and LinkedIn survey, 75% of knowledge workers were using AI at work, and 78% of AI users were bringing AI tools to work that their company hadn’t provided. More than half of users also said they were reluctant to admit using AI, for fear of looking lazy or replaceable. [3]

And in a PagerDuty survey released in June 2026, 66% of 1,250 non-IT office workers at large companies said they had used AI tools while knowing their company didn’t allow them. More than 30% reportedly had entered confidential information into public models. [8]

For anyone in security, this is a nightmare.

Contracts, customer data, source code, meeting notes and unannounced business plans get fed into AI that sits outside the company’s oversight. Nobody knows who gave what to which model. Even if results seem to be coming in, intellectual property and personal data may be leaking out behind the scenes.

But simply banning it doesn’t solve the problem either.

If the official tools can’t keep up with what people on the ground need, a ban doesn’t stop the use. It only makes it invisible.

---

GAFAM is part of this split too

This is where the story comes back to GAFAM.

Microsoft, Google and Amazon are trying to make enterprise AI their next big platform business. They are on the selling side: tying AI to a company’s identity systems, email, documents, cloud and access controls, and selling “official AI you can use safely.”

Meanwhile, consumer AI, starting with ChatGPT and Gemini, reached individuals without waiting for any company’s adoption decision. You can sign up in minutes, and the models update quickly. People on the front lines discovered how convenient AI is before their companies wrote any rules.

In other words, in today’s AI workplace there’s a big gap between management and individuals in how fast AI is being adopted.

Management moves while weighing contracts, security, company-wide rollout and return on investment.

Individuals pick whichever AI is easiest to use right now, to get today’s deadline done.

The Nanto side holds up the enterprise AI that GAFAM sells as the banner of organizational transformation.

The Hokuto side uses the latest models outside the company’s control to boost personal productivity.

For GAFAM, too, this isn’t simply a race over model performance. How much of the AI that employees use on their own can be brought back into a company-approved environment? Can access, data and auditing be managed without sacrificing convenience? That is the real battleground of the enterprise AI market.

It’s workplace design, not AI, that creates the two schools

Nanto and Hokuto aren’t a difference in personality.

They are reactions to the incentives a company creates.

At a company where announcing that you used AI earns you credit, more people will use AI even on work that doesn’t need it.

At a company where announcing that AI cut your workload means your job, or your colleagues’, gets cut, more people will hide their results.

At a company that bans AI while demanding only the productivity gains from it, shadow AI will spread.

All three grow out of the same problem.

Nobody has decided who receives the benefits of adopting AI, and who takes responsibility when it fails.

That said, there’s no simple answer to how the benefits should be shared.

If you hand the freed-up time straight to the person who made the efficiency gain, it may not lead to improvements for the whole team. If you tie results directly to pay, people may chase only the numbers that are easy to measure. And if the company takes all the gains for itself, employees will start hiding how they improved things.

Amazon’s KiroRank presumably started out as a way to spread AI use and find good practices. The moment it turned usage volume into a ranking, employees optimized for token counts rather than customer value. Even Amazon, a company that provides AI, couldn’t figure out on the first try what to measure to encourage the right behavior.

What’s needed isn’t turning everyone into Nanto, or rooting out Hokuto. In fact, these two schools aren’t really fighting each other at all.

Measure results by customer value, quality, time, rework and profit, not by how many times people used AI. Give part of the slack created by efficiency back to employees as pay, rest, learning or new challenges. Before punishing someone for using unapproved AI, ask why the official tools weren’t good enough to get the job done. And turn the personal workflows that succeed into organizational knowledge without putting the person who built them at a disadvantage.

Of course, companies will still get things wrong. AI and work are both changing fast, and an evaluation method that works today may not be right six months from now.

That’s exactly why, rather than designing the perfect system in one go, what matters is a workplace where employees can say honestly, “This AI isn’t useful,” “Actually, I’m using my personal account,” or “I made things more efficient and only got more work.” Management and the front lines need to keep searching for the right way to use AI together, without hiding failures and adjusting the metrics as they go. Without that dialogue, Nanto will dress up the numbers and Hokuto will go underground.

A company where people can use AI openly is strong.

But that doesn’t mean a company that forces people to use AI.

It means a company where you can be honest about using it without losing your job or your standing.

🦉 Come to think of it, I hear that among the 108 branches of Nanto Seiken, there’s one modeled on the owl: Nanto Yakyō Ken, the “South Dipper Night Owl Fist”…

---

References

[1] TechCrunch: Shopify CEO tells teams to consider using AI before growing headcount

[2] Amazon: Message from CEO Andy Jassy: Some thoughts on Generative AI

[3] Microsoft WorkLab: AI at Work Is Here. Now Comes the Hard Part

[4] Glean Work AI Institute: Work AI Index 2026

[5] BetterUp Labs: Workslop is the new busywork. And it’s costing millions

[6] Mint: Amazon tells staff to stop using AI just to use AI after shutting down its token leaderboard

[7] PNAS: Evidence of a social evaluation penalty for using AI

[8] PagerDuty: Report Finds Two-Thirds (66%) of Office Professionals Have Used Unauthorized AI Tools at Work