A curated survey of 94 prominent engineers describing exactly how they use coding agents. Each entry names the agents and model families they favor, summarizes how code review is handled in practice, and gives one concrete practice or ritual the engineer recommends adopting. The dataset was assembled from public repos, talks and posts and links each claim to its original evidence; the stated aim is to show what actually works so readers can pick one technique to try in their own workflow.
Clear patterns emerge: common tooling includes Claude Code, Codex/OpenCode, Amp, Cursor and provider-specific CLIs, with models ranging from Claude Opus/Fable to GPT-6/Astra and niche models like Grok. Review strategies split between automated cross-model or adversarial reviews, spot-checking high-risk areas, and full human line-by-line checks for core paths; many teams limit agent output to what they can realistically review. Practical rituals crop up repeatedly - autogenerated pre-commit reviews, subagent pipelines, QA passes per release, fresh-eyes reviewer bots, daily code limits, sensors/sidecars pointing to risky diffs, and agent-written tests or screenshots on PRs. The bottom line: agents are broadly embraced but tightly scaffolded with review patterns, CI loops and minimal-risk rules so automation scales without giving up human control.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
QEMU's Jazz emulator can now boot and run Windows NT 3.1 on MIPS architecture with full high-resolution graphics and networking. A ready-to-use image and source code are available for users to try this setup. (virtuallyfun.com)
Deslop is a set of rules designed to improve AI-generated TypeScript code, based on Oxlint guidelines. It aims to enhance code quality and consistency in AI-assisted development. (github.com)
Anthropic developed an in-house version control system called Scion to handle a tenfold increase in daily code merges, reaching over 30,000 per day. This new system supports their large-scale AI research and development processes, replacing Git due to the size and complexity of their monorepo. (anthropic.com)
Kyle Kukshtel discusses integrating large language models into game development to improve efficiency and iteration. He emphasizes practical, ongoing use of AI tools like Claude Code for real-world projects, beyond just demos. (kylekukshtel.com)
Standard C has many incompatibilities with how C is implemented in real-world compilers like gcc and clang, leading to non-conformance by default. These differences include handling of asm keywords, floating-point rounding modes, and support for standard pragmas, which are often ignored or unsupported by compilers. (sebsite.pw)
Sqlthon is a lightweight Python package that simplifies SQLite3 database operations. It provides a straightforward interface for managing SQLite databases efficiently. (github.com)
Today's best Hacker News stories, summarized and screenshotted, one email a day.