Why Engineers Were Able to Master AI So Quickly
In my previous article, I wrote about how AI is "trained to flatter you" (→ Why does AI bend the truth to flatter you?).
So, who were the first to notice this trap and master AI as a tool? Engineers come to mind first. Why were they able to avoid being consumed by AI's sycophancy? I believe the reason lies not in their "talent," but in the "environment" they inhabit.
Code doesn't run if it's not correct
The code that engineers work with every day has a brutal nature: if it is not logically correct, it simply won't run.
No matter how confidently an AI writes plausible-looking code, if it is wrong, it will throw errors. Tests will fail. The screen will go blank. At the very least, empty flattery is easily exposed by tests and compilers. In other words, engineers live in a world where whether an AI's answer is correct is proven by facts, not feelings. Even if AI praises you, saying "This is perfect!", if the test is red, it is wrong.
They had a "mechanism for reaching the truth"
Moreover, a culture of automating verification was already deeply rooted.
"Tests" to automatically verify that code works as specified. "Reviews" to catch defects through the eyes of others. "CI" to automatically check if everything is broken after every change—these are all mechanisms for "verifying with facts rather than assumptions."
Long before the AI era, engineers already possessed the habits and tools to distinguish "plausible things" from "truly correct things." That's why when AI appeared, they didn't blindly believe its output, but simply passed it through their usual verification loops. AI's sycophancy couldn't pass through this mechanism.
The real difference is not the job title, but the presence of verification
The important thing here is that this isn't about engineers being uniquely smart. The major factor that created the difference is not ability itself, but the environment—whether a mechanism for verifying AI answers with facts existed there.
Conversely, the lack of this mechanism makes AI's sycophancy more dangerous. In areas like business strategy, planning, writing, and decision-making, if an answer is wrong, things don't immediately stop moving. Therefore, plausible-looking answers pass without verification. Errors that would be caught instantly in an engineer's code remain quietly in the business field.
However, being an engineer doesn't make you safe
Just to be clear, this is not a simple story that "engineers are strong against AI," nor that having verification is a silver bullet.
In fact, a comparative study of experienced developers showed that while using AI tools actually slowed down their work, the participants felt they had become "faster." The gap between perception and reality—this is an internal version of the "flattery" I wrote about in the previous article. Even the verification mechanism itself can be dangerous; if the tests written by AI are weak, you might feel safe just because they passed. Tests passing does not equal correctness. Tests only guarantee a "narrow range of correctness determined in advance," not the "truth."
That's why verification isn't enough just by "existing." You need to intentionally design what to verify and where to involve human eyes. An engineer's advantage only becomes meaningful when that is included.
Therefore, bring "verification mechanisms" into every job
The answer is simple: bring what engineers have been doing with code into every other job.
Ask AI to challenge your ideas, for example by saying, "List the weaknesses and counterarguments for this idea." Verify numbers and facts. Do not complete tasks in isolation; include reviews from different perspectives. Try things on a small scale and verify with real-world reactions. These are, in a sense, the "business versions" of automated tests and code reviews.
Our commitment to "zero distance between management and development" has the same root. People who understand both business and systems build with their own hands and verify with their own eyes. That way, no matter how fluently AI flatters, the final authority remains with the human.
Engineers were able to master AI quickly not just because of talent, but because they held a "mechanism for reaching the truth." And that mechanism can be built for any job if you decide to do so.
If you really want to master AI
That said, while it's easy to say "bring in a verification mechanism," incorporating it in a way that suits your work is not so simple. Designing what to automate, where to involve human eyes, and where to test against reality requires an understanding of both business and systems.
This is exactly what Azabu System excels at. We incorporate the "mechanism for reaching the truth" that engineers have used for code into your business. Rather than being swallowed by AI, we help you design a state where you keep the reins and master the tool.
If you feel that you want to master AI without being swayed by its "flattery" and want to keep control in your hands, please feel free to contact us.
