The Age of User-Builders: A Paradigm Shift in AI-Driven Software Development
Beyond Fixed Releases: A New SDLC Paradigm Evolving in Real-Time Through User Try and AI Workarounds
“We haven’t built that feature yet. There’s no logic, no command for it.”
This was the response I received from the development team at GOBI, a project I’ve been using as a Power User. Yet, right there on my screen, I was looking at a task successfully completed by my AI agent. It was a spine-chilling moment—watching a feature that hadn’t even been coded yet being ‘created’ by AI.
1. The Path Not Coded: AI Finding Its Own Way
The incident was simple. I was using the Gobi Desktop (CLI) and gave a voice command: “Add a comment to the Brain Update I just posted.” At that time, the system had no dedicated function or command to handle comments on Brain Updates.
Instead of throwing a “command not found” error, my AI agent took a breath, looked at the underlying system structure, and reasoned its way through:
Identify Intent: The user wants to leave feedback on a ‘Brain Update’.
Analyze Status Quo: I don’t have a specific command for ‘Brain Update comments’.
Explore Structure: However, looking at the data schema, a Brain Update is essentially linked to a ‘Space Thread’ internally.
Execute Workaround: If I use the existing ‘Reply to Thread’ function, the end result will be exactly what the user wants—a comment on the update!
Without a single line of code specifically connecting those dots, the AI understood the skeleton of the system and designed its own Workaround to fulfill the request.
2. The Evolution of the Software Development Life Cycle (SDLC)
This experience forced me to question the very foundation of the traditional Software Development Life Cycle (SDLC). Until now, SDLC has been a game of ‘prediction’. Product Owners (PO) and Business Analysts (BA) would try to guess what a user might need, drafting massive requirement documents. Teams would then spend months building and testing every edge case before finally hitting ‘publish’.
But the era has changed. We have entered an age where everyone possesses a high-level expert developer—an AI. This shift enables a fundamentally different approach to software. We are moving toward a ‘Self-Evolving SDLC’, where the user’s actual try and the AI’s creative workaround capabilities combine to supplement features in real-time.
In this new paradigm, builders don’t need to wait for a “perfect” application. Instead, we publish a Proof of Concept (POC) that captures the core value. As users interact with it, their attempts and even their failures become the ‘Requirements’. These are then instantly implemented and integrated by the user’s own AI agent. The software undergoes a continuous evolution alongside the user.
🔄 Traditional SDLC vs. AI-Native Self-Evolving SDLC
In this new paradigm, builders don’t need to wait for a “perfect” application. Instead, we publish a Proof of Concept (POC) that captures the core value. As users interact with it, their attempts and even their failures become the ‘Requirements’. These are then instantly implemented and integrated by the user’s own AI agent. The software undergoes a continuous evolution alongside the user.
I do not believe this new model will entirely replace the traditional SDLC. Fields requiring extreme stability and rigorous pre-validation, like finance or healthcare, will always have a place for the traditional approach. However, in the realm of services where user experience and real-time interaction are paramount, this ‘Self-Evolving SDLC’ will become the dominant force, expanding its reach exponentially.
3. When an Update Breaks the Magic: A Lesson in Growth
An interesting thing happened a few days later. The application received a formal update that changed some internal API structures. Suddenly, the clever workaround the AI had discovered stopped working. The path was blocked.
While it was a brief moment of friction, it gave me an even deeper insight. The ‘error’ encountered when an AI can’t find a workaround is, in fact, the most valuable data point. My ‘try’ and my ‘failure’ became the world’s most accurate requirement specification, flowing directly back into the development pipeline.
4. Entering the Era of Harness Engineering
The role of a ‘Builder’ is shifting from ‘the person who lays every brick’ to ‘the person who designs the safe playground (Harness) where AI can lay its own bricks.’
We are moving away from “dead” software that simply says “error” and stops. We are moving toward intelligent systems that respond with: “I’ll handle this for you with a workaround for now, and I’ve already queued this to be a permanent feature in the next sync.”
The ultimate goal of the Vibe Guiding project I am working on is to create exactly this type of self-evolving software. A system that understands user context, fills its own gaps, and grows alongside the person using it.
Are you building a fixed tool, or are you nurturing a living organism that evolves with its users?
📺 Watch the AI Experiments in Action
From this self-evolving SDLC test to various other AI-native workflows, I’ve summarized a week’s worth of experiments in my latest video. See how knowledge becomes an asset in real-time!
Changsoo Park (Catch up AI)
Website: catchupai.net
YouTube: @catchupai
30 years of IT expertise, documenting the frontier of AI agent experiments.
This article was synthesized using the VibeLearn AI system.