Have an LLM build a new language toolchain
A developer got Claude to produce the compiler, runtime, and tests for 5emel while barely inspecting the code themselves.
- What they did
- They spent a few weeks prompting Claude Opus 5.5 to assemble the tooling for 5emel, an experimental language of their own. The project grew into roughly 15,000 lines of TypeScript/JavaScript plus 3,000 lines of C. They said they scarcely opened the code, including the compiler, and instead used long experience with programming books to keep the project on track.
- How it went
- The end result was a functioning toolchain. When another Claude instance was shown one function and a description of most of the language, it could describe a lot of it, but not every detail.
- Tips
- This kind of project still leaves you with plenty to untangle yourself. Build in short passes, and pause before any step that would permanently alter files, outputs, or generated code.
Try it yourself with Claude
I want to create a small experimental language called [language name]. Its main rules are [feature 1], [feature 2], and [feature 3]. Please help me design the language, then generate a compiler or interpreter and a runtime in [TypeScript/Python/C]. Break the work into small stages, explain what each stage changes, and check the examples after every stage. If a step would delete files, overwrite major modules, or otherwise make an irreversible change, stop first and ask for my approval. Finish by summarizing the system design and listing any parts that need extra review.
Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.
Get 3 agent uses like this every morning
With the day's top AI news. Free, and one click to unsubscribe.
Related agent uses
Scan commits for security bugs before they land
Use Claude Code to inspect each commit for secrets, injection flaws, and migration risks before it reaches the repository.
Tighten homelab access without locking yourself out
One person used Claude Opus to harden a homelab, but it cut off its own access and stopped; OpenAI’s Sol then completed the changes.
Use spare AI quota to draft reviewable PRs
A founder pointed an agent at goals and task lists so it could turn leftover Claude and Codex quota into pull requests for review.
Use Copilot to scaffold a full-stack app
They let GitHub Copilot start a new full-stack app, then kept fixing setup issues until the React interface was styled and usable.
Use Claude across many branches for app work
A developer spread work across multiple Claude chats, then used adb and logs to let Claude test app behavior on a phone.
Use Copilot agent mode to finish a web app
Jonas used GitHub Copilot with Claude Sonnet to turn a game-score idea into a working app.