Note: Unless otherwise specified, "Trae" in this article refers to the international version with the Trae Solo feature enabled.
Prologue: From Tools to Mindset—The "Role Shift" of Developers in the AI Era
Hackathons have always been arenas for battling over technical skills and stamina. It wasn't until I won third place at the ByteDance Trae Solo Hackathon that I truly realized: AI has completely reconstructed the development paradigm, even changing the very identity of a "developer."
This competition was a rehearsal for the transition from a "code craftsman" to an "AI collaboration director." I hope this review can help every builder using AI to construct the future find their next leverage point.
1. The True Value of AI Isn't Making You Write Faster, but Daring to "Unlearn"
a) Creative Kickstart: Abandoning Inertia and Letting AI Lead the Way
The night before the competition, I had GPT-5 thinking brainstorm some mini-hackathon ideas for me. I ultimately chose this direction:
Ritual Radio | Emotional Radio Hook: Use a "gesture slide" to draw a line representing your mood for the day, and the system generates a 5-minute audio and font animation. This serves fragmented moments like commuting or before bed, making self-regulation a ritual.
The inspiration came from my daily use of Endel (a white noise app) and my obsession with multi-sensory interaction. But what's truly important isn't the idea itself, but the courage to let AI participate in "defining the problem," no longer getting bogged down in whether I could write all the code myself.
b) Unlearning Old Paradigms, Actively "Giving Up Control"
Throughout the process, I barely wrote a single line of code. Instead, I adopted "document-driven development":
- First, I used Gemini 2.5 Pro to break down the requirements and output the core functional plan (not the code implementation).
- I fed all key information like "technical roadmap" and "interaction abstraction" to the AI, focusing only on whether the MVP could be launched via the shortest path, completely avoiding getting stuck on details like UI, interaction, or color schemes.
- I used the Trae Solo feature to let the AI generate a prototype directly. When the AI didn't understand something correctly, I reflected on my prompt design or the model's limitations, rather than "burying my head in the sand."
This is what I understand as "unlearning" in the AI era: actively letting go of the old habit of "having to do everything yourself" and shifting your energy to defining problems and mastering the essence of the experience.
2. AI Is Not Just a "Tool," but a "Virtual Partner"
a) Multi-Agent Collaboration—Directing Your AI Team Like a Film Director
During the entire development process, I used more than one large model:
- Gemini was responsible for planning and breaking down tasks. It's meticulous, supports multimodality, and has a large context window.
- Claude focused on code implementation. It has strong execution abilities and is suitable for following a set plan.
- GPT-5 thinking was used to break through creative bottlenecks and "redefine" product interaction and experience.
- By the way, I'm still more accustomed to using o3. I used it to plan my hackathon schedule for the day. It performed the best and even told me what I should be doing at different times.
Each model is like a "partner" with a different personality and specialty. Only when you can assemble them into a cohesive team can AI help you accelerate your implementation tenfold.
b) A New Division of Labor—Humans Focus on "Definition + Aesthetics + Curation"
You'll find that the more powerful AI becomes, the more humans need to:
- Define the product's soul (What is the core hook of the experience? How do you make the user feel the "ritual"?)
- Guard the aesthetics (AI can handle color palettes, animations, and sound effects, but the final cut still needs a human touch.)
- Curate and integrate resources (Stitching together different models, tools, and inspirations into a complete experience.)
In this process, your identity is no longer that of a "technical worker," but a "Prompt Director + Experience Designer."
3. From "Minimal Input" to "High-Density Experience": A New Proposition for Product Philosophy in the AI Era
a) The Design Insight Behind the Minimalist Slider
My final project had only a three-dimensional slider (Energy/Brightness/Speed) and a breathing font animation. But this "extremely simple" interaction allowed the judges to "play with it" on the spot and feel the shift in atmosphere within 10 seconds. This actually points to a trend in current AI product design:
In an age of fragmented attention, allow users to enter a sense of ritual through the shortest path, creating a personalized emotional space with minimal interaction.
b) "Stitching" Is Not Opportunism, but a New Form of Creativity
You could say I was "stitching": throwing together Endel, community animations from v0.app, and AI-generated color schemes. But this stitching is actually the maximization of information density and emotional density. AI gives builders unprecedented space for trial and error and transformation. The key is whether you can "define a clear line" and let the model reassemble it for you.
4. Debriefing the Debrief: What Is the Ultimate Value of an AI Hackathon?
- The hardest part isn't "doing it fast," but "doing it right": Only by breaking down the problem into small, clear pieces can AI help you assemble the final product like an assembly line.
- AI is not a "code monkey," but a "team of partners": You have to learn to orchestrate, not just prompt.
- Unlearn, define the experience, and focus on humanity—this is the final moat for humans in the AI era.