[About 80% of Instructors See Increased OJT Burden] Rookie Engineers' Use of Generative AI Burdens the Field: What are the Surfacing Pitfalls of Engineer Training?
A survey by Job Support reveals that while 90% of rookie engineers use generative AI, it has paradoxically increased the OJT burden in 80% of workplaces due to a lack of fundamental understanding and problem-solving skills.
📋 Article Processing Timeline
- 📰 Published: April 22, 2026 at 19:00
- 🔍 Collected: April 23, 2026 at 00:02 (5h 2m after Published)
- 🤖 AI Analyzed: April 23, 2026 at 00:32 (29 min after Collected)
Job Support Co., Ltd. (Location: Chiyoda-ku, Tokyo, Representative Director: Tsutomu Tsukada) conducted a survey on "Generative AI Usage by Rookie Engineers and the Burden of Guidance" targeting those who have experienced educating and guiding (OJT) new graduates and young engineers (1st to 3rd year at the company) within the past two years.
Survey Overview: Survey on "Generative AI Usage by Rookie Engineers and the Burden of Guidance"
[Survey Period] March 16 (Mon) to March 17 (Tue), 2026
[Survey Method] Internet survey via PRIZMA
[Number of Respondents] 1,004
[Survey Target] Monitors who answered that they had experienced educating and guiding (OJT) new graduates and young engineers within the past two years at the time of answering the survey
[Survey Source] Job Support Co., Ltd.
[Monitor Provider] Sacrisa
In engineer training, the use of generative AI has become an inevitable trend.
However, looking at OJT in the development field, concerns have been raised that issues such as "lack of self-driving ability" and "insufficient retention of basic knowledge," which have been pointed out previously, have surfaced more prominently due to the spread of generative AI. What is the true nature of "self-driving ability" that is truly required in the education of rookie engineers in the generative AI era?
Detailed data including answers such as "Skills needed to master AI," "Appropriate learning order," and "Reducing the burden by improving OJT" are available in the white paper.
90% answered that rookie engineers use generative AI in their work! What are the issues with the code they create?
First, when asked, "Which is closest to the generative AI usage situation of the rookie/young engineers you are in charge of in their work?", combining "Actively utilizing (40.0%)" and "Utilizing as needed (50.0%)" shows that 90% in total use generative AI on a daily basis.
Under such circumstances, what kind of issues are arising with the code created by rookie engineers using generative AI in actual work?
From here, we asked those who answered "Actively utilizing" or "Utilizing as needed" in the previous question.
When asked about "Issues faced regarding the code created by rookie engineers using generative AI," the most common answer was "The person themselves does not understand the mechanism or basis of the output code (61.4%)," followed by "Unable to decipher the requirements and giving vague instructions (prompts) (47.5%)" and "Unable to independently identify and fix the cause of errors (36.6%)."
It can be read that a "fundamental lack of understanding" such as the basis of the output code is an issue in the field.
In addition, giving vague instructions without accurately grasping the requirements and being unable to independently identify and repair the cause of an error were also raised as issues.
Furthermore, when asked about "The 'reaction/response' seen in rookie engineers when there is a bug or error in the AI-generated code," the most common answer was "Repeatedly asking AI blindly without reading the error message themselves (46.5%)," followed by "Throwing the basis of the answer to AI, saying 'I don't know because it's AI output' (45.8%)" and "Immediately asking seniors for the correct answer without investigating the cause themselves (40.5%)."
It is thought that having a convenient tool at hand eliminates the behavior of thinking logically and verifying oneself, resulting in an environment where error resolution skills are hard to foster.
Due to the spread of generative AI, the OJT burden has increased in about 80% of workplaces. The causes of the guidance burden are the lack of "sense of ownership" and "basic knowledge."
How do these behavioral processes of rookie engineers change the actual guidance system and the burden on senior engineers?
When asked, "How has the 'time required for guidance and code reviews (OJT burden)' changed since rookie engineers started using generative AI?", the following results were obtained.
"Significantly increased (increased effort for rework and retraining from basics) (26.1%)"
"Slightly increased (51.8%)"
"Unchanged (18.1%)"
"Slightly decreased (3.3%)"
"Significantly decreased (guidance became easier) (0.7%)"
About 80% answered that "the guidance burden has increased" since rookie engineers started using generative AI, suggesting a situation where new educational costs have arisen to verify AI outputs.
In such a situation, what is the fundamental reason why the use of generative AI does not lead to "reducing the burden" of guidance?
Survey Overview: Survey on "Generative AI Usage by Rookie Engineers and the Burden of Guidance"
[Survey Period] March 16 (Mon) to March 17 (Tue), 2026
[Survey Method] Internet survey via PRIZMA
[Number of Respondents] 1,004
[Survey Target] Monitors who answered that they had experienced educating and guiding (OJT) new graduates and young engineers within the past two years at the time of answering the survey
[Survey Source] Job Support Co., Ltd.
[Monitor Provider] Sacrisa
In engineer training, the use of generative AI has become an inevitable trend.
However, looking at OJT in the development field, concerns have been raised that issues such as "lack of self-driving ability" and "insufficient retention of basic knowledge," which have been pointed out previously, have surfaced more prominently due to the spread of generative AI. What is the true nature of "self-driving ability" that is truly required in the education of rookie engineers in the generative AI era?
Detailed data including answers such as "Skills needed to master AI," "Appropriate learning order," and "Reducing the burden by improving OJT" are available in the white paper.
90% answered that rookie engineers use generative AI in their work! What are the issues with the code they create?
First, when asked, "Which is closest to the generative AI usage situation of the rookie/young engineers you are in charge of in their work?", combining "Actively utilizing (40.0%)" and "Utilizing as needed (50.0%)" shows that 90% in total use generative AI on a daily basis.
Under such circumstances, what kind of issues are arising with the code created by rookie engineers using generative AI in actual work?
From here, we asked those who answered "Actively utilizing" or "Utilizing as needed" in the previous question.
When asked about "Issues faced regarding the code created by rookie engineers using generative AI," the most common answer was "The person themselves does not understand the mechanism or basis of the output code (61.4%)," followed by "Unable to decipher the requirements and giving vague instructions (prompts) (47.5%)" and "Unable to independently identify and fix the cause of errors (36.6%)."
It can be read that a "fundamental lack of understanding" such as the basis of the output code is an issue in the field.
In addition, giving vague instructions without accurately grasping the requirements and being unable to independently identify and repair the cause of an error were also raised as issues.
Furthermore, when asked about "The 'reaction/response' seen in rookie engineers when there is a bug or error in the AI-generated code," the most common answer was "Repeatedly asking AI blindly without reading the error message themselves (46.5%)," followed by "Throwing the basis of the answer to AI, saying 'I don't know because it's AI output' (45.8%)" and "Immediately asking seniors for the correct answer without investigating the cause themselves (40.5%)."
It is thought that having a convenient tool at hand eliminates the behavior of thinking logically and verifying oneself, resulting in an environment where error resolution skills are hard to foster.
Due to the spread of generative AI, the OJT burden has increased in about 80% of workplaces. The causes of the guidance burden are the lack of "sense of ownership" and "basic knowledge."
How do these behavioral processes of rookie engineers change the actual guidance system and the burden on senior engineers?
When asked, "How has the 'time required for guidance and code reviews (OJT burden)' changed since rookie engineers started using generative AI?", the following results were obtained.
"Significantly increased (increased effort for rework and retraining from basics) (26.1%)"
"Slightly increased (51.8%)"
"Unchanged (18.1%)"
"Slightly decreased (3.3%)"
"Significantly decreased (guidance became easier) (0.7%)"
About 80% answered that "the guidance burden has increased" since rookie engineers started using generative AI, suggesting a situation where new educational costs have arisen to verify AI outputs.
In such a situation, what is the fundamental reason why the use of generative AI does not lead to "reducing the burden" of guidance?