Will AI Replace Me?
Which kinds of work will AI replace? From standardized tasks and the impact on engineers and creative professionals to whether companies really need AI and the risks of adopting it, this article presents my view of the current AI wave.
Almost every month or quarter brings a new AI model. As we watch these models solve increasingly complex problems, it is difficult not to be impressed by their rapid progress—and equally difficult not to wonder: Will AI replace my job?
A few years ago, the technology industry was talking about cloud computing, followed by big data. Now the focus has shifted to AI. The difference is that this wave of AI is no longer an idea confined to movies or an internal technology initiative for businesses. It has already entered the workplace and everyday life.
This article presents my observations from several angles: what AI is likely to replace, what it is less likely to replace, whether companies genuinely need to adopt it, and how we should approach living and working alongside it.
Will AI Replace My Job?
The Short Answer: AI Is More Likely to Replace Tasks Than Entire Jobs
If a job can be broken down into a clear standard operating procedure, requires little additional judgment, and involves limited interpersonal or business accountability, many of its tasks are likely to be automated by AI.
By contrast, when work is highly customized and involves substantial expertise, business context, trade-offs, and accountability, AI is more likely to act as an assistant than a complete replacement—at least in the near term.
Instead of asking, “Will my profession disappear?”, a more practical question is:
Which parts of my daily work can already be described clearly, repeated consistently, and verified easily?
Those are usually the first parts to be automated.
Work That Can Be Clearly Defined Is the Easiest to Give to AI
Companies establish working hours, operating procedures, and approval rules because explicit standards make people and processes easier to manage. Defining, announcing, and implementing those rules takes effort, but once a standard exists, the same process can be repeated consistently.
We do something similar when using AI: we describe its capabilities, define the boundaries within which it may operate, and specify its inputs, outputs, and validation criteria. When a task can be described completely and AI can repeatedly produce acceptable results through the same process, it will naturally begin to replace the manual effort involved.
The hardest part is often not execution, but defining the problem, handling exceptions, and deciding who is responsible when something goes wrong.
Why Do Engineers and Creative Professionals Feel the Impact So Strongly?
Software development has a steep learning curve, but many entry-level tasks are highly repetitive, such as form-based CRUD work, data transformation, and bill of materials (BOM) explosion. In the past, once someone crossed the initial technical barrier, a basic skill set could remain useful for years. AI can now assist with many of these templated tasks. What is being disrupted may not be the profession of software engineering itself, but work that consists primarily of turning explicit specifications into predictable code.
Art and design face a different situation. Professional ability—whether in watercolor, sketching, or digital tools—usually takes years to develop, yet the final result may still fail to match a client’s expectations. When clients cannot clearly articulate what they want, some now ask AI to generate many options and select the one closest to what they had in mind. This lowers the cost of producing an initial draft and reduces some of the work that would previously have been outsourced.
But generating an image is not the same as completing a design, just as generating code is not the same as understanding a system. Clarifying requirements, judging quality, and accepting final responsibility still require people.
Do We Really Need AI?
AI is popular, but not every problem requires it. The market likes to discuss prompts, image generation, and AI agents because these topics are new, attract attention, and can easily be packaged into courses. Cloud architecture, programming fundamentals, and improvements to existing processes are less likely to command the same attention.
The value of a technology, however, does not come from how fashionable its name is. It comes from whether it can solve a problem at a reasonable cost.
This reminds me of a common lesson about big data: a company does not always need to build a massive analytics platform before it can improve revenue. Instead of beginning by analyzing every user action, drawing heat maps, designing complex product bundles, and spending heavily on advertising, it may be better to start with one clear and testable problem—for example, identifying customers who have not purchased for a long time, sending them a coupon or return incentive, and observing whether they come back.
There are real-world examples of this approach. In China, Guopi Mobile used restaurant transaction data to help merchants send targeted coupons to returning customers. A premium coffee chain used promotional offers to reactivate lapsed customers, increasing the share of active customers from 17% to 25%. Another restaurant loyalty campaign targeted less-active members with rewards and reported a 54% increase in visits and a 42% increase in monthly revenue. These cases used data at different levels of sophistication, but they shared one principle: each began with a specific business problem and tested the outcome through a measurable action.
This does not mean that big data has no value. It means that adoption should not begin with “Which technology should we use?” but with “Which problem are we trying to solve?” If a simple customer segment and coupon experiment can validate the direction, there is no need to begin by building a vast data platform. Data and model complexity can be added gradually when simpler methods are no longer sufficient.
The same applies to AI. If rules, search, automation scripts, or existing tools can complete a task, adding an AI layer will only introduce more cost and uncertainty. Whether to adopt AI should depend not on how popular it is, but on whether it is better suited to the problem than the alternatives.
What Are the Risks of Adopting AI?
The risks do not come only from the model itself, but also from how an organization uses it. When work is handed to AI without retaining the ability to verify results, transfer knowledge, and communicate clearly, a tool intended to improve efficiency can make problems harder to detect.
Results That Cannot Be Verified
The greatest risk is not that AI is occasionally wrong, but that we do not know how to determine whether it is right.
If users do not understand the domain in which AI is operating and have not defined success criteria, they will struggle to verify its output. It is like asking someone how to become a billionaire or achieve financial freedom: the answer may sound entirely reasonable, but the outcome depends on timing, resources, and circumstances. It is difficult to quantify and impossible to reproduce directly.
Before adopting AI, an organization should be able to answer at least the following questions:
- What specific problem is it supposed to solve?
- What happens if the output is wrong?
- Who verifies the output and accepts final responsibility?
- Is there a simpler and more reliable alternative?
If none of these questions has an answer, AI may not bring efficiency. It may simply replace a visible process with uncertainty that is harder to notice.
Gaps in Talent and Organizational Knowledge
As more entry-level work is handed to AI, a longer-term question emerges: where will the next generation of experts come from?
People usually begin with foundational work and gradually understand the business logic behind a system through implementation, mistakes, and debugging. If a company focuses only on the short-term efficiency of automation without redesigning how people learn, newcomers may lose the opportunities through which experience is built.
Similarly, if process and decision logic exists only in prompts or AI agents without being genuinely understood by anyone, the organization may gradually lose that knowledge. When regulations, requirements, or workflows change, the existing automation may continue to run, but no one may know where to adjust it or what else the change will affect.
Even after AI takes over foundational tasks, companies must deliberately preserve opportunities for learning, review, and knowledge transfer. Otherwise, they may gain speed today at the cost of their ability to maintain systems tomorrow.
Ambiguity in Language and Context
Consider French in comparison with other languages. French grammar often requires speakers to state subjects, tense, and referents explicitly, while Chinese and Japanese more frequently omit some information and rely on listeners to infer it from context.
This does not mean that French is free of ambiguity, or that any language is inherently more precise than another. Languages differ in how much information they encode directly in a sentence and how they depend on context. When the other party is an AI rather than a person, those differences can affect how requirements are interpreted.
Natural language is inherently ambiguous. The same sentence can carry different meanings depending on segmentation, context, and the situation in which it is spoken. Consider this Chinese phrase:
1
愛上一個人
It can be segmented as 愛上／一個人 (“fall in love with someone”) or deliberately read as 愛／上一個人 (“love the previous person”). The written characters are identical, but readers rely on context to determine the intended meaning.
The Japanese phrase 大丈夫です presents a similar problem. Depending on the context, it can mean “That’s fine” or serve as a polite way to say “No, thank you.” Reading only the words while ignoring tone and context can produce the opposite interpretation.
For example, suppose I have a Tokyo Subway Ticket and ask ChatGPT, “How should I get from Ueno to Ikebukuro while making full use of this ticket?” It will follow those conditions and recommend various routes covered by the ticket.
But if I ask a friend who knows me, “How should I get from Ueno to Ikebukuro on July 5, 2026?”, they may infer from the date, my interests, and the destination that I am going to Sunshine Creation 2026 Summer at Ikebukuro Sunshine City. Instead of planning a route only as far as Ikebukuro Station, they may suggest using Higashi-ikebukuro Station so that I arrive closer to the venue I actually want to reach.
We often believe that our requirements are already clear. Only after stating them do we discover that a friend, stakeholder, or client may understand what we actually want better than AI does, because they know the background and context we never explicitly provided.
After receiving a prompt, AI infers the user’s intent from the text and context available to it. When a requirement depends heavily on shared understanding, implication, or “reading the room” without stating the conditions and expected result clearly, AI can only choose an interpretation that appears plausible. It may complete the task fluently while producing something other than what the user truly wanted.
Adopting AI therefore involves more than turning something we would say to a person into a prompt. Hidden requirements must be converted into explicit conditions, examples, and acceptance criteria. Otherwise, AI does not eliminate the cost of communication; it merely postpones the misunderstanding until after the output has been produced.
How Should We Approach AI?
My answer is: accept it, embrace it, but do not follow it blindly.
AI can help us organize information, produce first drafts, explore alternatives, and accelerate tasks that once consumed significant time. But it still has limits, and the quality of its output depends on whether the user understands the problem, provides enough context, and can verify the result.
The most important skill is not simply knowing how to operate a particular AI tool. It is knowing when to use it, when not to use it, and how to maintain independent judgment about its output.
AI may replace some of the tasks we perform, but it does not necessarily replace people who can define problems, make trade-offs, and accept responsibility. Instead of repeatedly asking, “Will AI replace me?”, it may be more useful to examine the value we provide: are we merely repeating a process, or are we contributing understanding, judgment, and accountability that AI cannot exercise independently?
References
The results below come from media coverage and case studies published by service providers. They illustrate possible approaches but do not imply that every company will achieve the same results by adopting a similar strategy.