# Seven Weeks of Vibe-Coding with Claude: From Idea to a Working RSS Service
Journalist Thomas Claburn from The Register built a commercial RSS service called RSScal in seven weeks and 337 commits using Claude Code. Development cost: $40 for a Claude subscription plus $14 monthly for VPS. The stack includes Docker, Python with FastAPI, Celery, Redis, PostgreSQL via Supabase on the backend, and SvelteKit with Tailwind CSS on the frontend. The author emphasizes: without basic experience in RSS apps, prompting the model would have been ineffective.
Vibe-Coding as the New Standard
Claburn, a coder since the 1980s, previously built RSS readers by hand: the Electron app Vulture Feeds and its rework RSSputin. With the release of Anthropic Opus 4.5 and OpenAI Codex 5.2 at the end of 2025, vibe-coding has become routine. Models generate working code—imperfect, unoptimized, but functional.
Key challenge: the model is competent at generation but helpless with context. Claude mixed up production and dev builds, forgot rate limiting, but suggested unexpected design solutions. Developers have to balance trust and control.
Comparison with Other Cases
- Simon Willison vibe-coded a macOS app for presentations.
- Security researcher Michael Taggart: "AI worked, but the process was hated."
- Jim Nielsen failed with an RSS reader: generation is simple, but refinement, distribution, and maintenance—no.
These examples confirm: AI simplifies the start, but doesn't solve operational tasks.
Takeaways for the Market and Skills
Market implications:
Building a SaaS competitor in a month for $40 threatens template products and CodeCanyon freelancing. SaaSpocalypse is real for niche services.
Skill development:
The claim "you don't learn with AI" is a myth. Claburn improved his Docker, Python, and SvelteKit skills. Full delegation atrophies skills, but active vibe-coding accelerates learning.
Comparison to the 1970s: pro developers react like academic musicians to punk—outraged by the lack of mastery, but the result works.
Key Points
- Vibe-coding is effective for hobby projects: 337 commits in 7 weeks on a stack with no prior experience.
- Models generate working code but require basic knowledge for prompting and fixes.
- Threat to template SaaS and freelancing; niche services are vulnerable.
- Skills grow with active oversight; full automation is harmful.
- First commit: February 22, 2026.
— Editorial Team
No comments yet.