Claude Code Powers AI Agents to Build My Entire Side Project in Real Time

Claude Code’s agent‑driven workflow turned a one‑hour prototype into a publicly hosted Next.js app, demonstrating how Anthropic’s internal coding tool can move from side‑project to production without human‑written code. According to the author’s March 15 post on dev.to, the prompt given to Claude Code was deliberately minimal: “build a web app where I paste my resume as plain text on the left, see it formatted on the right, and can export to PDF.” Within minutes the AI generated a complete codebase, deployed it to Vercel’s free tier, and produced a functional UI that the author christened OhMyDoc. The resulting app runs entirely client‑side, with no backend database, no runtime AI calls, and a monthly cost of $0, as the author notes. This mirrors Bloomberg’s observation that Claude Code originated as a side project before becoming a core component of Anthropic’s product strategy, underscoring the tool’s capacity to bootstrap full‑stack applications from a single natural‑language instruction.

The technical architecture of OhMyDoc is deliberately simple. The app uses Next.js for server‑side rendering and Tailwind CSS for styling, both of which are standard choices for rapid front‑end development. Hosting on Vercel’s free tier eliminates any ongoing infrastructure expense, while the absence of a database means all processing occurs in the browser. The author emphasizes that the AI’s involvement ends at build time; the runtime logic relies on deterministic parsing of plain‑text input and CSS‑driven layout, ensuring that the same input always yields the same formatted output. This separation of concerns—AI for code generation, deterministic code for execution—aligns with Wired’s broader analysis of AI agents, which notes that recent web‑browsing agents from OpenAI, Anthropic, and Google are designed to complete tasks autonomously but still depend on human‑crafted runtime environments for reliability.

One of the most notable outcomes of the AI‑generated code is the inclusion of an XML editor, a feature the author did not request. The editor allows users to modify specific text nodes directly in the document structure, avoiding a second round of AI rewriting when minor phrasing changes are needed. The author describes this as the “surprise feature” that turned out to be the most useful part of the tool. The Decoder’s recent piece on AI coding agents highlights a similar pattern: simple text‑based inputs can yield sophisticated, unexpected capabilities when the underlying agent applies broader design heuristics. In this case, Claude Code inferred that users would want granular control over AI‑generated resume content, pre‑emptively embedding a low‑level editing interface that sidesteps the need for additional AI calls.

From a product‑market perspective, OhMyDoc directly addresses a pain point identified by the author: existing “free” resume formatters invariably require sign‑ups, impose paywalls, or rewrite content with proprietary models. By eliminating login, email capture, and any server‑side AI processing, the app offers a privacy‑first, truly free alternative. The author’s experience suggests that AI‑built tools can rapidly iterate to meet niche user needs without the overhead of traditional development cycles. However, the visual design, while competent, lacks distinct branding—a criticism echoed in the dev.to post, which notes that AI‑generated UIs often exhibit a generic “Tailwind energy” that may not differentiate a product in a crowded market.

The broader implication for the AI industry is the validation of Claude Code as a production‑ready code‑generation engine. Anthropic’s internal use of the tool, as reported by Bloomberg, indicates that the company sees strategic value in automating routine development tasks, freeing engineers to focus on higher‑level product work. If AI agents can reliably produce deployable, cost‑free applications like OhMyDoc, enterprises may begin to re‑evaluate the economics of low‑complexity software projects, potentially shifting a portion of the development labor market toward prompt engineering and agent supervision. The author’s open‑source release on GitHub provides a concrete case study for other developers to replicate, and the community response—inviting others to share similar AI‑built projects—suggests a nascent ecosystem forming around agent‑generated code.