“AI Coding is a Nightmare” — What’s Happening on the Ground

A post by an engineer on Hacker News has sparked a heated discussion.

“After several iterations, the codebase turned into a mountain of unnecessary code.”

This is a story about a project collapsing as a result of over-relying on AI coding assistants. Many voices resonated with this sentiment, with phrases like “AI coding is a nightmare,” “AI is boring,” and “Coding without AI is the new revolution” circulating within the engineering community.

If there are small and medium enterprise (SME) owners watching this trend and thinking, “Maybe we don’t need to use AI after all,” they should pause for a moment.

By accurately understanding the essence of this “AI fatigue,” opportunities for SMEs may actually come into view.

—

The Essence of AI Fatigue — Three Specific Symptoms

When we organize the complaints raised by engineers, the issues can be broadly categorized into three main points.

1. The Curse of Reinventing the Wheel — Creating the Same Function Multiple Times

AI cannot recognize that the same functionality already exists within the existing codebase. As a result, it generates the same logic repeatedly. The code bloats, making it difficult to discern which is the original and which is a copy.

In one project, it was reported that simply removing duplicate code generated by AI reduced the code volume by 40%. This means that 40% of the code “written” by AI was unnecessary from the start.

2. Lack of Overall Design — Missing the Forest for the Trees

AI focuses solely on the task at hand. If you ask it to “write this function,” it will do so. However, it does not consider how that change will ripple through the entire system.

A human engineer would intuitively understand, “If I change this, that will break over there.” AI lacks this understanding. As a result, every time one area is fixed, another area breaks, creating a chain reaction of fixes.

3. The Boredom of “Everyone is Making the Same Thing”

This is where the phrase “AI is boring” comes from. The code generated by AI is an average of the training data. It simply returns the most frequently occurring patterns, leading to similar structures across different projects.

For engineers, coding is inherently a creative act. The moment it transforms into a task of “inspecting AI output,” motivation collapses.

—

Is “Coding Without AI” Really a Revolution?

In reaction to these sentiments, the claim has emerged that “coding without AI is the true revolution.”

I understand the feeling. However, we should take a calm look at the situation.

This discussion pertains to senior engineers in Silicon Valley, earning between 20 to 50 million yen annually. They have a comprehensive understanding of the architecture in their minds and can write high-quality code quickly without AI. Thus, they can say, “AI is a hindrance.”

What about the situation in SMEs?

There are no dedicated engineers. Outsourcing a modification of internal systems costs between 500,000 to 2 million yen. Even simple data aggregation automation is often done manually using Excel.

We should not discuss these two worlds on the same playing field.

The draft cited data claiming that “development costs could be reduced by 30% without using AI,” but the source is unclear. In reality, the opposite is true; a survey by GitHub found that task completion speed improved by up to 55% when using AI assistants. Of course, it depends on the conditions, but it’s not a simple matter of “removing AI will lower costs.”

—

Three Reasons SMEs Should Continue to Use AI

The “AI fatigue” within the engineering community accurately reflects the limitations of AI. However, if used with an understanding of these limitations, AI remains a powerful weapon for SMEs.

Reason 1: AI Can Fill the Gap of “No Experts”

The biggest challenge for SMEs is the lack of specialized personnel.

For example, modifying a website can cost between 50,000 to 200,000 yen when outsourced, with a two-week turnaround. Using an AI coding assistant, an employee with some knowledge of HTML might be able to handle it in 30 minutes. If there are ten modifications in a year, that could eliminate outsourcing costs of 500,000 to 2 million yen.

The key point is not to use AI as a substitute for senior engineers but rather as a tool to fill the gaps where there are no experts. The evaluation changes when the purpose differs.

Reason 2: Breaking Dependency on Individuals

The most frightening situation for SMEs is being in a state where “things cannot operate without that person.”

The aggregation logic known only by Mr. Tanaka in accounting. The customer response manual that only exists in Mr. Sato’s head in sales. Such tacit knowledge can be transformed into a system using AI.

Specifically, feeding business procedures into AI to create manuals, translating decision criteria into flowcharts, and placing inquiry responses onto AI chatbots can significantly reduce the risk of dependency on individuals with tools costing a few thousand to tens of thousands of yen per month.

The cost of leaving a state where “if Mr. Tanaka leaves, it’s over” versus the monthly fee of 10,000 yen for an AI tool is clear.

Reason 3: The Cost of “Experimentation” Has Dramatically Decreased

This is the most important structural change.

Previously, creating a prototype for a new service could cost 3 million yen and take three months when outsourced. Now, using an AI coding assistant, there are cases where a working prototype can be created for under 50,000 yen in 1 to 2 weeks.

At 3 million yen, “failure is not an option.” At 50,000 yen, “let’s give it a try.”

This difference is decisive. One of the few points where SMEs can compete with large corporations is “speed of decision-making,” but if the cost of experimentation is high, that speed cannot be leveraged. By reducing experimentation costs to one-sixtieth, AI finally makes the agility of SMEs a weapon.

—

So, What Should We Do?

The essence of “AI fatigue” lies in the misuse of AI.

When top engineers in Silicon Valley say, “AI is a nightmare,” it’s because for them, AI is like a “low-skilled pair programmer.” If someone less skilled than you is chiming in, it’s certainly a hindrance.

However, for SMEs, AI is a “better helper than having no one at all.” It doesn’t have to be perfect. If it can produce a score of 70, humans can fix the remaining 30. It’s far more rational to let AI work for 50,000 yen and finish it yourself than to pay 2 million yen for outsourcing.

Here are three specific actions:

1. Don’t Rely Entirely on AI. Accept that AI output is a “draft.” The final judgment should be made by humans.
2. Start Small. Rather than immediately integrating AI into core systems, begin with creating internal manuals or simple data aggregation.
3. Decide What Not to Use AI For. Avoid using AI in areas where security is critical or where there are legal risks. Defining boundaries for where to use AI and where not to use it will determine the success of AI utilization.

—

AI Fatigue is a “Correct Learning Process”

When the internet first emerged, people said, “There’s too much information to use effectively.” When smartphones appeared, they were criticized for blurring the boundaries between work and personal life. Both were valid points, but companies that discarded the technology were eliminated.

AI is no different. The current “AI fatigue” is merely a stage of learning about the limitations of the technology. Companies that understand these limitations and find ways to use AI that suit their needs will prevail.

What I want to convey to SME owners is not to be swayed by the complaints of Silicon Valley engineers. The challenges they face are fundamentally different from ours.

Their challenge is “AI lowering the quality of their work.” Our challenge is “not having enough people, money, or time.”

AI remains overwhelmingly effective against the latter challenge.

The fatigue comes from misusing it. Learn correctly, fatigue correctly, and use it correctly. That’s all that’s needed.