I got buried in repetitive code

At 10 p.m. last Tuesday, I was in a rented apartment in Hangzhou, updating a booking page for a client. Buttons, modals, API calls all looked similar. The real time sink wasn’t the ideas; it was repeatedly checking files, copying old logic, and making sure I hadn’t broken anything. I’ll admit it plainly: I knew how to write the code—I was just too slow at it, and both my rhythm and attention were worn down.

What DreamCoder is, and who’s using it

DreamCoder is an AI coding editor created by the author of Mastering Emacs. Its core capability is understanding project context, then using natural language to help generate or modify code. Developer Kyle Simpson has used it to explore new ways of programming. But from several discussions I saw, it’s more of a copilot: it doesn’t own your architecture or deployment decisions. I also tried it on a small client utility. When I described “add filters to the order page” in natural language, it produced a code draft I could keep editing, saving me some research time.

What it costs to replicate today

The software has a free trial; current pricing depends on the official website. My rough starting budget was 0 yuan to try it, plus 30–60 minutes to learn the editor and one clearly defined task in an existing small project. The technical threshold is basic: I need to read simple code and understand my own business logic. The first step isn’t exploring every feature; it’s opening the site, going to the download page, installing the editor, creating a practice project, and asking the AI to generate the simplest possible input form. I made this mistake before: I treated “it can generate code” as “it can deliver software,” then spent hours changing someone else’s basic assumptions. Now I review the output first, then verify it section by section.

Where I’d use it at different stages

If I were just starting without stable clients, I wouldn’t buy a course or overcomplicate my setup. I’d use it on one practice page and see whether it reduces my typing time. It’s also fine not to try it yet; writing down the requirement clearly may matter more. With one or two clients, I’d test it on a real, repetitive feature, record the time before and after, and then decide whether to continue. While scaling, I’d treat it as a drafting tool before a teammate writes code, while keeping code review, testing, and deployment ownership explicit. This tool isn’t for everyone: if I never want to touch code and only want to drag and drop a page together, a simpler site-building service may fit better. If I tried it, I’d start with a small task that doesn’t affect an existing project.