Navigating AI Development: Building Software While the Rules Keep Changing

Last week, I released Protocollie, a project built in just four days using languages I don't fully understand, without directly writing much of the code myself. People often ask how I managed to achieve this, but the truth is, I'm not entirely sure it can be replicated. We're all navigating this emerging AI landscape as we go, making it up as we encounter new challenges and solutions. There's a common phase in new technologies where everyone claims expertise, but with AI, we're in a unique period. The rules are constantly evolving, making it impossible to claim true mastery. The most seasoned AI pair programmers have been in the field for only about two years, which means we're all perpetual beginners. This rapid pace of change disrupts any established methods and forces us to adapt continuously. My "system" for building Protocollie evolved organically rather than through deliberate planning. It started with a single document to remind myself of the architecture I discussed with Claude, an AI assistant. As the project progressed, I added more documents to streamline the process: Architecture Overview: Initially a README, it described the basic structure of the project. Technical Considerations: A collection of detailed notes on issues Claude encountered and solutions we devised. Workflow Process: Steps I noticed were repeated frequently, documented for consistency. Story Breakdown: Project tasks broken into 15-30 minute chunks to keep the AI on track and manage forgetfulness. These documents weren't created because four is the optimal number, but because it's the point where I stopped adding more. The system felt effective, but its permanence is questionable. Tomorrow, something else might work better, and that's the beauty and challenge of this era. Building Protocollie was an unusual experience. On that first Sunday, I spent about 90 minutes giving directions, testing features, and providing feedback. The rest of the time, I lived my life while Claude worked in the background. It was like having a hyper-efficient, tireless junior developer. This shift in the input-to-output ratio felt surreal, breaking my traditional mental models of productivity. There are moments when I feel guilty about this new approach. It seems too easy, almost like cheating. But this is the reality of modern AI development—rapid iteration and continuous experimentation. Our processes are less about following established procedures and more about throwing spaghetti at the wall and seeing what works. Sometimes, it's the act of throwing rather than the sticking that matters most. Each era of technology introduces new abstraction layers, simplifying complex tasks. From hand-carved HTML tables to sophisticated AI assistants, the tools evolve, but the essence of creativity and problem-solving remains. In the AI age, the key skills are not syntax or algorithms; they're coherent desire, precise imagination, and structured wishing. My documents didn't focus on code but on memory, frustration management, and patterns that emerged during the project. We're all junior developers again, not in the traditional sense where experience brings expertise, but in a permanent state where the technology advances faster than knowledge can accumulate. It's akin to being a surfer in an ocean that constantly changes its physics—just as you think you understand the waves, they shift in unpredictable ways. This can be both terrifying and liberating, depending on your comfort with control. The four-document system isn't a prescriptive template but a snapshot of one person's experience in a rapidly evolving field. I've made these documents available on GitHub, not as gospel but as a potential starting point. They represent a specific week in 2025, a moment frozen in time. Tomorrow, someone else might discover a different system that works for them—a three-document method, a five-document method, or perhaps none at all. The goal isn't to find a perfect process but to embrace the dynamic nature of AI development. We're all building sandcastles at low tide, knowing the water's coming back, and enjoying the process while it lasts. Industry Evaluation Industry insiders view this shift as a pivotal moment in software development. The ability to leverage AI for rapid prototyping and development is transforming how teams operate. Companies like Meta and Google are investing heavily in AI-driven tools and platforms to stay competitive. Scale AI, a leader in data annotation, recently secured a significant investment from Meta, underscoring the importance of quality data in training advanced AI models. Wang's move to Meta highlights the strategic importance of integrating AI expertise across various roles within companies. His departure and the appointment of Jason Droege as interim CEO suggest that collaboration and flexibility will be crucial in this rapidly evolving field. The key takeaway is that while traditional methodologies are valuable, the future of software development will require a willingness to experiment and adapt, leveraging AI's capabilities to enhance productivity and creativity.