Retrospective: Building a Side Project with AI, and What Comes After

 

Looking Back on a 20-Part Journey

"I want to build a webpage that does tarot readings." One sentence. That's where this project started. Designing data for 78 cards. Setting the mood with dark navy backgrounds and gold text. Creating animations where cards float as if breathing. Solving dozens of mobile issues. Twenty installments documenting the entire process.

In this final part, I want to step back and view the project from a high altitude. The moments AI genuinely helped and the moments it didn't. The collaboration methodology this project crystallized. And the fundamental value of building a side project with AI.

The Full Project Timeline

Here's the Tarot Master development process in chronological order. Idea refinement and tech stack selection: one day. Card data and interpretation text for 78 cards: two days. Core reading functionality and UI: three days. Animations and micro-interactions: two days. Real-time AI interpretation and serverless API: two days. PWA conversion and user features: three days. Mobile optimization and UI overhaul: four days.

Total: roughly two and a half weeks. This wasn't full-time focus but evenings and weekends, so actual hours invested were less. Without AI, the same result would have taken at least two months. That time difference decides whether a side project lives or dies. A two-month side project has an extremely low completion rate, but two and a half weeks is finishable before your energy runs out.

Five Things AI Did Well

First, domain knowledge transfer. I knew almost nothing about tarot. AI enabled me to quickly grasp the meanings of 78 cards, various spread layouts, and the difference between upright and reversed readings. It dramatically flattened the initial learning curve when entering an unfamiliar domain.

Second, bulk content generation. Writing upright and reversed interpretations for each of 78 cards, a total of 156 text blocks, would take a person a week. With AI, it was done in two days. The texts weren't used as-is; they went through review and editing. But having a draft to work from versus starting from a blank page is a world of difference.

Third, UI implementation acceleration. Color palette recommendations, CSS animation code, responsive layout patterns. On a solo side project without a designer, having AI provide both creative direction and implementation code simultaneously was decisive.

Fourth, starting points for problem-solving. Chrome mobile's bottom bar bug, service worker caching issues, jspdf Korean font problems. For each specific technical issue, AI suggested multiple approaches. The final fix required hands-on testing, but having a starting point versus starting a Google search from zero is a significant efficiency gain.

Fifth, SEO strategy. AI assisted with keyword targeting for the blog series, meta tag structure, and content architecture design. Technical content SEO is hard to approach without experience, and AI provided concrete guidelines.

Four Things AI Couldn't Do

First, final judgment. "Should I add this feature or not?" "Should I fix this bug now or build the new feature first?" "What impression will this design give users?" At every decision point about the project's direction, AI could present options but never bore the responsibility of the final call. When asked "Which is better, A or B?" AI lists the pros and cons of each, but it can't say with conviction "A is right for this project."

Second, sensory validation. "Does this animation speed feel natural?" "Does this color combination actually create a mystical atmosphere?" "Is this button position comfortable for thumb tapping?" AI cannot make these sensory judgments. The difference between 0.3 and 0.4 seconds, between #0a0a1a and #0a0a2a, can only be assessed by a human looking at a screen.

Third, real-device testing. The Chrome mobile bug was the prime example. AI proposes theoretically correct solutions, but whether they actually work on a physical device is unknowable without testing. Verifying behavior across different devices, browsers, and network conditions is work AI cannot perform.

Fourth, creating the "why." As discussed in Part 19, establishing a project's reason for existence and direction is something AI cannot do. "Why build a tarot web app?" The answer came from personal interest, technical curiosity, and passion for side projects -- not from AI.

Five Principles of AI Collaboration

Here are five principles distilled from the 20-part series.

One: a vague start is fine. You don't need a perfect spec. A single sentence of an idea is enough. AI breaks that vagueness into concrete decision axes. Lowering the barrier to starting is the first value of AI collaboration.

Two: I design the structure, AI handles execution. Architecture, data models, component structure -- the developer leads design. After the design is set, AI rapidly writes the code. This division of labor is the key to achieving both quality and speed.

Three: never proceed without understanding. Read and understand every piece of AI-generated code. Ask immediately about anything unclear. This process feels slow, but it saves ten times the time during future expansion and debugging.

Four: always ask "why." Don't accept AI's first suggestion at face value. Confirm why this approach was chosen, what alternatives exist, what the tradeoffs are. These questions simultaneously produce better outcomes and deeper understanding.

Five: validate sensory experience personally. Test AI's output in real environments. Design, animation timing, touch responsiveness, overall usage flow. Never skip the validation that only eyes and fingertips can perform.

The Greatest Advantage of AI Collaboration for Side Projects

Among many advantages, if I had to name just one: reducing startup friction.

The most common reason side projects die is failure to start. Energy drains during the planning phase, or you hit a wall in an unfamiliar domain, or decision paralysis over tech stack selection. AI dramatically reduces all of this initial friction.

Capturing the energy of the moment an idea sparks and immediately connecting it to concrete form. That's the most fundamental value AI collaboration brings to side projects. Tarot Master became an actual working product instead of staying an "idea" precisely because of this value.

Balancing Speed and Quality

Working with AI is fast. But fast isn't always good. Chasing speed makes it easy to sacrifice quality. "It works, good enough" was a temptation at every turn.

Consciously maintaining that balance was important. Build fast but verify carefully. Accept AI's suggestions but don't follow blindly. Copy code but understand it first. Maintaining that tension is itself a development competency for the AI era.

Looking back, the most satisfying parts of Tarot Master aren't the features that were built quickly. They're the parts where I invested time in details. The screen transition tuned to exactly 0.3 seconds. The color testing that led to selecting #0a0a1a. The dozens of touch tests repeated on mobile. No matter how fast AI generates code, the time required for these details can't be compressed.

Expansion Possibilities: Stories Yet Untold

Tarot Master is closer to a starting point than a finished product. There's still so much I want to explore. More sophisticated AI interpretations that adapt to the user's specific question context. Additional spreads beyond the Celtic Cross. Weekly and monthly insight reports analyzing daily readings. Multi-language support to serve users worldwide.

These expansion possibilities are the fuel that keeps a side project alive. Seeing a next step rather than an endpoint. The more you build, the more you want to build. Tarot Master became that kind of project.

"Building Together"

As I close this 20-part series, here's what I want to say: AI doesn't build it for you. You build it together.

This distinction matters because as AI tools advance rapidly, the perception spreads that AI "does it for you." Toss it a prompt and magic happens. For simple things, that's true. But for projects you care about deeply, projects meant to give users a meaningful experience, "for you" isn't accurate. "Together" is.

I made every decision in Tarot Master myself. Choosing the navy background. Deciding to build free-choice mode. Overhauling the UI to Linear's style. AI helped me realize those decisions quickly. That relationship is what healthy AI collaboration looks like.

Time spent writing code decreased, but time spent thinking actually increased. With AI shouldering the execution burden, I could focus more on "what to build" and "why to build it." And that focus is what made Tarot Master not just "a tarot app" but "my tarot app."

To Everyone Who Read This Series

Thank you for joining me on this long journey from Part 1 through Part 20. I hope that this process -- starting from a single sentence and arriving at PWA, mobile optimization, real-time AI interpretation, and beyond -- gives someone out there the small spark of courage to think "Maybe I should try this too."

Side projects ultimately belong to the people who start them. Imperfect beginnings, not perfect preparation, are what bring projects to life. AI is the best tool for supporting those imperfect beginnings. But pressing the start button is still up to you.

If there's an idea floating around in your head, throw one sentence at an AI right now. That one sentence might become a 20-part story.