One week. One engineer. One AI coding assistant. A Cloudflare engineer reportedly rebuilt Next.js - the React framework that powers a substantial portion of the modern web - using that combination, and the writeup is a useful look at what AI-assisted development actually delivers.
What Actually Got Built
Rebuilding all of Next.js would be impossible in a week. The full framework is years of accumulated work: routing, server-side rendering (generating HTML on the server before sending it to the browser), image optimization, API routes, and a compilation pipeline. What the engineer built was a functional implementation of Next.js's core runtime behavior - specifically the parts needed to run Next.js applications on Cloudflare Workers, their serverless computing platform.
Runtime compatibility work - making a framework designed for traditional Node.js servers run correctly in a different environment - is notoriously tedious. It requires reading through thousands of lines of implementation code, identifying what the target environment needs, and writing shims (small code patches that fill gaps) for everything that doesn't transfer directly. That kind of scoped rebuild would normally take a team multiple weeks. One engineer did it in one.
Where AI Tools Actually Deliver
This case study is worth examining because it identifies exactly where AI coding assistance has the most impact: rebuilding something that already exists.
When there's a reference implementation - software whose expected behavior you understand and are trying to replicate - AI coding tools move fast. Tools like Cursor and GitHub Copilot can hold context across multiple files, generate large working code blocks, and check output against a spec. The engineer still reviews and fixes mistakes, but the gap between understanding what needs to be built and having a working first pass collapses significantly.
Greenfield product work with fuzzy requirements is a different story. If you can't precisely define what "done" looks like, AI tools can't reliably build toward it. But "make Next.js run on Cloudflare Workers" is a well-specified problem with a clear finish line. That specificity is what makes one-engineer-one-week possible.
For developers evaluating AI coding tools, this kind of case study is more informative than abstract productivity percentages. The productivity multiple is real - but it's highest on well-defined, bounded tasks, not on open-ended creative engineering.