6 months and $485: My journey into building with AI (harshdeepgupta.substack.com)

What started as an “overnight” idea turned into a six-month, $485 journey to ship PicPickr, a swipe-based photo picker for newlyweds. The author leaned heavily on AI tools—Cursor to prototype, Lovable to generate UI designs, Supabase for backend, Resend for email flows, and Electron (with Forge) to handle local file access and packaging—plus Claude Opus to untangle a thorny build config. Major surprises included scaling local filesystem handling for thousands of large photos, the invisible work of packaging/signing binaries, CI/testing, and the landing-page and email-integration work; roughly 500 prompts and a lot of manual wiring later, a beta is live at picpickr.com. For the AI/ML community this is a useful reality check: generative tools can dramatically lower barriers and speed iteration, but they don’t remove systems engineering, integration, and distribution complexity. AI-generated UIs (Lovable) and code accelerators got the project moving, yet integrating outputs into a real codebase, managing configs, packaging apps, and handling large local data remain time-consuming. The takeaway: AI is a powerful bridge to ship prototypes, but expect significant non-AI work—momentum and orchestration matter more than raw model output.