Note: The text for this article was fully structured and proofread by Gemini 3.1 Pro.
Hello.
Since December 2025, the performance of generative AI has been improving at a terrifyingly fast pace. Many of you might be feeling the anxiety that writing code by hand will soon become virtually obsolete, and that you might be left behind in this rapidly accelerating world.
Amidst this, I participated in a long-term programming contest (under a different account name).
Interestingly, the rule for this contest was “Generative AI usage is allowed”.
What happens when everyone fully leverages AI to compete seriously at a time when its performance is skyrocketing? I was very curious to find out.
In this post, I want to share the insights I gained from that experience and my future strategy for personal development.
What Happened at the AI-Allowed Contest
To get straight to the conclusion: There was almost no difference in rank fluctuations between when everyone used AI and when everyone wrote code by hand.
This was a highly fascinating result.
Of course, the overall capability of all participants was significantly boosted by AI. If I compare my current AI-assisted code with the code I used to write by hand back in the day, the current code is undoubtedly and overwhelmingly stronger.
However, the relative skill gaps and rank changes were not that pronounced. Observing the timeline of the contest, the following trends emerged:
- Overwhelmingly Fast Start With AI, basic logic and initial implementations are finished in an instant.
- Improvements Hit a Ceiling Based on Know-how However, subsequent improvements (finding bottlenecks or fundamental changes to the algorithm) heavily depend on the participant’s own distinct know-how and contest experience.
“Peripheral Tasks” to Boost Scores Kept Increasing
So, after everyone finished their initial implementations, what were they doing? They were asking the AI to handle “peripheral tasks” to secure more points.
For example:
- Asking the AI to build peripheral tools like visualizers and testing frameworks.
- Getting the AI to write parameter-tuning scripts.
- Setting up automated adjustments and parallel execution environments.
Personally, whenever I programmed as a hobby in the past, I usually avoided tasks like “writing tests” or “setting up error handling and CI/CD” because they were tedious. However, thanks to generative AI, these tasks can now be done in almost no time, which means I find myself “accidentally doing them.”
As a result of everyone’s baseline skill being raised, the reality of an environment where everyone can use AI was this: “The skill gap remains unfilled, while the sheer volume of everyone’s peripheral work (like tool creation) simply keeps increasing.”
As AI performance will undoubtedly continue to rise, we might never again experience a contest under these exact transitional conditions. It was a truly valuable experience.
Escaping the Tsunami: My Future Development Strategy
Based on these experiences, I’ve thought about my personal “survival strategy for development” going forward.
In short, it is to “escape from the tsunami that is generative AI.”
Currently, generative AI is wreaking havoc in the realm of “systems programming”—areas that can be completed entirely through terminal inputs and outputs or CI pipelines. This is an environment suited to AI’s strengths: a highly logical, closed world, which is already saturated with highly skilled individuals.
Because of this, I intend to deliberately step out of that arena.
Specifically, I plan to flee towards areas that absolutely require a human’s “touch” and feedback on “whether it is actually fun,” such as YouTube video production and Roblox game development.
The fun factor in entertainment and gaming fundamentally relies on “manual work” and “human intuition”—building prototypes, having actual humans play them, and making continuous adjustments. Consequently, this high volume of hands-on work naturally leads to token savings (AI API usage fees and request limits). But above all, I inherently enjoy the entertainment field.
From now on, rather than focusing solely on writing perfect, lightning-fast code, the true survival strategy seems to be knowing “where to deploy AI and where to leverage human sensibility.”
Thank you very much for reading!