A senior engineer recounts stopping hand-written coding after 18 years because modern code-generation agents became faster and more productive than manual work. At Filestage, adopting AI tools (various models including Claude Code's Opus) with a $20 monthly subscription per developer raised merged PRs from roughly 200 to 300 per month without a sustained increase in bug reports. That extra throughput exposed infrastructure bottlenecks - Cloudflare tunnel limits and shared DB connection strain - which were resolved via infrastructure-as-code and open-source FRP tunnels. Rigorous CI guardrails (linting, type checking, duplication checks, 100% test coverage, end-to-end tests) plus AI code review are now integral to the workflow, and the author considers AI reviews indispensable for safe merging.
The piece argues against the idea that agents will default to generating assembly or microcode, insisting on a pipeline of LLM → high-level representation → deterministic compiler → machine code. High-level code preserves context, reduces the explosion of low-level details, and encodes intent through types and modules in ways assembly cannot, while compilers efficiently handle routine translation. The future envisioned replaces brittle text-file workflows with living, model-backed interfaces: UML/ER-style views, zoomable logic with production telemetry, and the ability to drill down to code only when necessary. The overall tone is practical and optimistic about a rapid, concrete shift in how software is built.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
The Hare compiler was ported to OpenBSD by modifying its FreeBSD support files and addressing build issues such as incompatible functions and format strings. Despite successfully building, the test suite failed due to OpenBSD-specific ELF binary requirements. (briancallahan.net)
Typst 0.15 introduces support for variable fonts, multi-file output, multiple bibliographies, and various PDF standards, enhancing document flexibility. The update also improves documentation, diagnostics, and migration tools to streamline user experience. (typst.app)
DHH states that agents implemented and optimized the Campfire web app in Elixir, Go, and Rust, with Ruby being the slowest. He emphasizes that productivity and developer joy outweighed speed concerns, but notes that maintainability is important when reading code becomes difficult. (twitter.com)
Build with Python is a beginner course where learners create drawings and interactive projects using Python, including making a robot and click-to-draw paintings. The course covers fundamental programming concepts like functions, variables, loops, and event handling. (scimigo.com)
Software engineering is shifting from coding to product engineering, focusing on solving user needs rather than writing code. Advances in AI and abstraction tools enable more people to build and use software without deep technical knowledge. (newsletter.chainofthought.show)
JetBrains developed a RAG pipeline called Air Context to improve semantic code search, enabling LLM agents to find precise, evidence-based code snippets from repositories. Semantic search allows agents to understand code meaning beyond keyword matching, increasing efficiency in large codebases. (blog.jetbrains.com)
Today's best Hacker News stories, summarized and screenshotted, one email a day.