How to use Claude to prototype a systems language
The author reports a core systems-language prototype in two weeks, followed by context and supervision problems that led them back to manual coding.
news.ycombinator.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Help prototype [systems language] with [core language features] and [backend target]. Maintain a memory bank of architecture and module relationships. Before changing a module, inspect its callers and dependencies and explain effects on the overall design for my review.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- khaledh
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The author built the language’s core pieces in two weeks but later abandoned the agent-heavy approach because of supervision burden and lost learning opportunities.
- Human involvement
- The author documented the project, repeatedly corrected cross-module reasoning and ultimately returned to manual coding.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Built core language and backend components with Claude Code.
- Documented the growing project using a memory bank.
- Repeatedly reminded Claude to inspect relationships with other modules.
An experienced software engineer used Claude Code for a hobby project: a new systems programming language built from scratch, including its backend. They report completing its core pieces in two weeks and estimate they could not have done the same alone in six months. The language was not fully fledged.
The implemented pieces included a static type system, algebraic data types, pattern matching, ownership and region-based memory management. The backend lowered code to SSA-based intermediate representation and included register allocation and x86-64 assembly generation. While developing it, the author documented the project using a memory bank so later Claude Code sessions could recover context without a fresh explanation.
Despite that documentation, they found Claude increasingly lost the overall design and pursued local optimizations that harmed the larger system. The author repeatedly reminded it to examine other modules and eventually became exhausted by that supervision. They also felt they had missed learning the topics deeply. They returned to writing code themselves with limited Cursor Tab help.
Who it's for: experienced developers evaluating rapid language prototypes and the supervision required as interconnected compiler code grows.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · khaledh
Frequently asked questions
Did the memory bank prevent context problems?
No. The author says Claude still lost the big picture and optimized locally, requiring repeated reminders to examine other modules.
Was the language complete?
The author explicitly says it was not fully fledged, although it included core language, memory-management and code-generation components.
More details
Remaining steps
- Returned to manual coding with limited Cursor Tab assistance.
About this account
The coding agents supported implementation while the author directed the project and supplied context.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=44740290. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: khaledh on Hacker News
Source published
Added to Agents In Use
Source checked
Tools and topics
Claude Code · Memory bank · Cursor Tab
Agentic coding · Development
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.