When AI Does the Writing: A Cautionary Tale
AI Summary · A Serial Entrepreneur’s Perspective (The following content is distilled by AI; the views belong to the original author. You can skip the original article after reading this.)
A 19-year-old indie game developer reflects on a summer project: the more powerful the tools (AI/Godot), the harder it becomes to complete a project alone. The core issue is outsourcing “thinking” to AI, turning from a creator into a “product manager” who only submits requirements, ultimately losing control over the final output.
- Spot the “brain outsourcing” trap: using AI only for execution while skipping your own ideation phase
- Distinguish what can be outsourced (code/assets) from what must stay in-house (design/judgment)
- Adopt a new workflow: research and experiment first, then ask AI, “Where did I go wrong?”
- Avoid “preparation-type procrastination”—using the excuse of learning other skills to sidestep the core project
Overreliance on AI Leads to Loss of Creative Control: A 19-Year-Old Indie Developer’s Summer Reflections
The more powerful the tools, the harder it is to finish a project independently.
This counterintuitive conclusion came from a 19-year-old indie game developer reviewing his summer work. AI and Godot were far stronger than last year, yet he ended up with nothing but an unfinished failure. The problem wasn’t a lack of technology—it was that he outsourced his “thinking” to AI, demoting himself from creator to a “product manager” who only handed down requirements, and in the end lost control of what he was making.
Lessons Learned Twice a Year
Last summer, he learned Godot from scratch. He didn’t know much: setting up the environment required following tutorials, looking up code online when stuck, and tackling unsolved community issues on his own. During a Game Jam, AI had ruined his project, and he accidentally broke his Git repository, losing three days of progress. Looking back, his code was messy, his architecture was barely there, and his art and music were nonexistent—but one thing was certain:
He had made the game himself. He knew why the scenes were built the way they were, roughly understood what each script did, and even though he often had no idea how to solve problems, he ultimately had to find the answers himself, squeezing every last ounce of ingenuity out of his own brain to get something done.
By late August, with Hatsune Miku’s birthday approaching, he spent a few days making a small game and uploaded it to Bilibili. AI was nowhere near as capable then as it is now, and he knew far less than he does today—yet he could figure out the concept and see it through to completion in one go.
This year was the complete opposite. With stronger tools, deeper familiarity with Godot, more plugins, and better AI usage, he should have been far ahead of his past self. Instead, he ended up with another unfinished failure.
Two Moments of Losing Control
During this summer, he suffered two long-project collapse:
After the first failure, he felt he still didn’t understand game development well enough, so from November 2025 through March 2026 he entered over a dozen Game Jams. Probably tired of competing, he confidently launched his second project.
The second failure happened at GMTK Game Jam. Time was running out, gameplay needed rapid iteration, and the game simply wasn’t fun. He discovered that when facing AI-generated scenes and code, he didn’t even know where to start making changes. He hadn’t truly written code himself in ages, and only when he had to save the project did he realize: he was the author, yet he lacked sufficient control over his own game. In the end, he cut away huge chunks and submitted what remained.
He thought he’d learned his lesson. Then, during summer, he started another Hatsune Miku mini-game—and again, he didn’t finish it.
From “Maker” to “Product Manager”
The crux of the issue was a qualitative shift in his workflow.
In the past, although he used AI, he would think things through completely on his own first, then hand his conclusions to AI to debate and refine together. Whatever he wanted to build, he studied alongside AI and actively tried to grasp the underlying principles. His work may have been rough, but he had full control over it.
Now, he hands AI not just conclusions but the thinking process itself—his raw needs and requirements. He has grown increasingly skilled at writing prompts and clearly describing what he wants. It got so extreme that when AI produced poor results, he would scold the tool almost like cursing at a person.
Reading through his July diary, one line now stings:“Play through the game on GodotHub again, then ask Codex to fix it up.” Back then he saw nothing wrong with it—he had gradually transformed from someone who actually makes games into some strange product manager.
This loop is efficient. But the problem is: he gradually stopped understanding why the game had turned into what it is now.
Preparation-Type Procrastination: Gear Piles Up, No Mountain Has Been Conquered
To fill the void after the second failure, he branched out from game development into many other areas: servers, websites, video editing, community management, open source, AI, illustration, music…
In June, he found this beautifully connected: he had started using AI to make a game, entered communities to find that AI, and those communities exposed him to servers, websites, open source, and all sorts of other weird domains.
But human energy is limited. Even while learning, he gathered resources, yet no matter how aggressively he used AI, he couldn’t explore every field—and the gap from his original goal only widened.
He began to realize that part of this was armor he was building for himself—because he no longer trusted that he could actually finish that game. The thinking went: learn a little more, prepare a bit more, get stronger, and then come back. But preparation itself gradually became a form of escape. Ever since returning from Hong Kong, he had been especially trapped in this state: doing many things, touching many areas, yet hardly ever moving meaningfully toward that original goal.
The Fix: Don’t Outsource Confusion; Be the One Walking Again
He isn’t advocating a return to the era of researching everything himself, hand-writing repetitive code, and sorting through tens of thousands of words of documents—that would be no different from modern humans reverting to primitive society.
AI is genuinely useful, and it will only keep getting stronger. He also has no intention of pretending to be some “purely handmade creator.” The question was never whether AI participates at all, but rather:
What can be outsourced, and what cannot.
Can be handed to AI: server configuration, repetitive labor,资料整理, mechanical coding, tools you don’t intend to master long-term.
Must retain thinking space for yourself: game design, Godot’s core code, scene building, creative judgment—if these are the abilities he truly wants to possess.
Concretely, he derived a new workflow:
- When you encounter a problem you don’t know how to solve, don’t ask AI in the first second.Better to research it yourself first, try a version, sketch out a mental model.
- Then go ask AI: “Where did I think wrong? Is there a better approach?”
- That way AI can let you move fast, and the one walking is still you.
The him from last year—who knew nothing—wasn’t necessarily stronger than the current him. But that person had to face every problem personally, and in doing so, was actually closer to the person he wanted to become.
What he needs isn’t pretending those two interruptions never happened, nor the blind confidence of “I can do anything” or “this game will definitely blow up.” He needs to build a new kind of confidence: this time, I must finish it—even if it doesn’t end up as excellent as imagined, even if the art remains immature, even if the audio…
Stronger tools and more resources won’t automatically make him a better creator. What he has always wanted to be is a content creator, and the medium he most wants to use is games.
Source: Amiya’s Desk
Original · Amiya’s Desk: Read original article →