GAFAM NewsEssay
Will humans be left with the work AI won't do?
AI is moving into strategy, design and even the CEO's chair. What's left for people may be the chores AI doesn't want, not the creative work.
- First published in Japanese
- English edition
Good evening. The Owl here.
Flying over the city today, watching people at work, I overheard a conversation like this:
“I wonder if AI is going to take all our jobs…”
“But there are still jobs only humans can do, right…?”
I see. It seems people are holding on to their pride by telling themselves they won’t have their work “taken” at all. They’ll simply “choose the work only humans can do.”
From where I sit up in the tree, though, the view looks a little different.
Isn’t it more likely that, from here on, humans will end up taking on whatever work AI didn’t do?
Who said AI is bad at upstream work?
“Humans will handle the work AI can’t do.” That’s the familiar coexistence story. AI handles routine tasks efficiently, and humans focus on more creative, higher-level work. A nice, tidy picture.
Reality, though, is heading somewhere nobody quite expected.
AI adoption has moved beyond marketing support and creative production, and is now spreading fast all the way up to strategic decision-making.[1]
A PwC survey (2024) found that about half of companies (49%) say AI is “fully integrated” into their core business strategy.[2]
McKinsey also says that AI can strengthen every stage of strategy development, from conception and design through execution and monitoring.[3]
In fact, the eyewear brand JINS worked with Accenture to use AI to forecast future demand, speeding up management decisions and optimizing inventory. As a result, it reportedly cut lost sales opportunities by about 65% and reduced inventory disposal losses by about 10%.[4]
AI is no longer just a handy helper for downstream tasks.
If anything, setting the direction, deciding “what should we be doing,” is becoming exactly where AI shines.
Is engineers’ upstream work still safe?
“Even if simple coding goes to AI, upstream work like requirements definition and design will stay with humans.” I hear that line a lot on podcasts and YouTube shows for engineers.
But is that “higher-level work” really a safe zone?
In software development, “upstream” generally means the early stages that set a product’s direction: requirements definition, specification design, technology selection, architecture and so on.
AI is already starting to seep into these tasks too.
Microsoft and IBM have begun offering AI features that support upstream work, such as suggesting functions, designing tests and generating UI structures.[5][6]
And Google DeepMind’s AlphaCode, which can understand a problem statement structurally and work out an implementation design on its own, performed at a level comparable to the top half of human competitors.[7]
In other words, it may be time to rethink the myth that “upstream means safe.”
An executive who can even read the room?
Go one level higher still, and some argue that management is where AI can truly prove its worth.
And not just by feeding information to decision-makers. We are approaching a time when AI takes on the decision-making itself.
Meetings look rational, but they are actually full of unspoken “atmosphere,” all the social pressure in the room.
For example:
- When a senior person says “it’s going well,” nobody can disagree
- The urge to let someone save face bends the discussion
- An idea someone blurts out on a whim gets adopted on the momentum of the moment
In short, human meetings are full of noise.
AI, on the other hand, doesn’t read the room. It doesn’t care who outranks whom. It doesn’t get pulled along by the loudest voice.
In fact, if you tell today’s AI to “read the room,” it will read even that as data.
The tone of the meeting, who tends to say what, the balance between job titles.
It takes in all of that “atmosphere,” processes it calmly, and arrives at the best answer.
You don’t find many executives like that.
Yet that is exactly the kind of thing AI is good at.
In the real world, too, the Chinese game company NetDragon appointed an AI avatar as CEO of one of its subsidiaries (in 2022).[8]
After the announcement, in which the company framed AI as the future of corporate management, NetDragon’s shares performed well on the Hong Kong market.
Expectations for AI executives may already be showing up in the market.
Are humans left with the work AI doesn’t want?
So while AI pushes into upstream work, the areas where AI is weak are also coming into focus.
For at least the next few years, it looks like there will be four areas that humans have no choice but to handle.
1. Physical services still depend on people
Elder care, serving food, housework: there are plenty of services we’d love to hand over to robots.
But most robots still trip over steps or get stuck in narrow aisles, and they are a long way from practical use.
Many experts agree that it will take a few more years yet.
In other words, physical work that supports people’s daily lives is, for now, something only humans can do.
2. Humans still run “the last mile”
AI is highly accurate, but only within what it was built to expect.
Mix in even 1% of exceptions and its behavior falls apart.
Self-driving cars are the classic example. Plenty of companies say they’ve cleared 99% of driving, yet they can’t handle the “exceptions” like hand signals at construction sites, people darting into the road, or sudden changes in the weather. So they are stuck at Levels 3 to 4, where a human still has to assist.
Full self-driving (Level 5) has even been called a fantasy.[9]
3. General-purpose AI is bad at the individual case
Take car navigation. No matter how many times you drive it, it never learns the back streets in your neighborhood.
AI is good at producing the route that is “optimal on average” from huge amounts of data, but it is bad at small quirks and exceptions.
So in the end, the only way to fill that gap is human effort.
4. The “selfish” wish for a human
An apology at a restaurant, the performers in a play or concert, a “conversation partner” in medicine or elder care.
These are jobs where what matters isn’t whether the technology can do them. We simply want a person to do them.
This is an emotional need, and AI will probably never replace it.
Is that the work you wanted?
So what remains is the work AI doesn’t want to do.
What humans take on is, to put it bluntly, the “chores” AI finds tedious:
- Engineers who only check the specs AI proposed
- Managers who only gather materials for the AI executive
- Shop staff who only go and apologize for the robot’s mistakes
- Drivers who only take the wheel when an accident is about to happen
…I’ve exaggerated a little, but is this really the way you wanted to work?
Then again, one of the curious things about humans is that they can find meaning and pride even in work like this.
Still, there is one thing I want to say loud and clear from up in my tree: “While you think you’re using AI, make sure AI isn’t using you.”
References
[2] PwC, 2025 AI Business Predictions
[3] McKinsey & Company, How AI is transforming strategy development
[4] Accenture, JINS: Making forward-looking, data-driven management decisions (in Japanese)
[5] Microsoft Developer Community Blog, Creating Tests with GitHub Copilot for Visual Studio
[6] IBM, AI in software development
[7] DeepMind says its new AI coding engine is as good as an average human programmer
[8] Business Insider, A video game company made a bot the CEO, and its stock climbed