Being a discreet human, not a jerk.
searchable musings on #Java, #Kotlin, Web, #Container & #Cloud, Code Quality, #InnerSource, #Security, #asciidoc #asciidoctor
| Identity Proof | https://bentolor.de/ |
| GitHub | https://github.com/bentolor/ |
"Across multiple coding agents and LLMs, we find that context files tend to reduce task success rates compared to providing no repository context, while also increasing inference cost by over 20%"
I've suspected this all along. Folks spending mucho-plenty time curating project-level .md files have been deluding themselves that it helps.
A widespread practice in software development is to tailor coding agents to repositories using context files, such as AGENTS.md, by either manually or automatically generating them. Although this practice is strongly encouraged by agent developers, there is currently no rigorous investigation into whether such context files are actually effective for real-world tasks. In this work, we study this question and evaluate coding agents' task completion performance in two complementary settings: established SWE-bench tasks from popular repositories, with LLM-generated context files following agent-developer recommendations, and a novel collection of issues from repositories containing developer-committed context files. Across multiple coding agents and LLMs, we find that context files tend to reduce task success rates compared to providing no repository context, while also increasing inference cost by over 20%. Behaviorally, both LLM-generated and developer-provided context files encourage broader exploration (e.g., more thorough testing and file traversal), and coding agents tend to respect their instructions. Ultimately, we conclude that unnecessary requirements from context files make tasks harder, and human-written context files should describe only minimal requirements.
Show HN: I built a sub-500ms latency voice agent from scratch (www.ntik.me)
Link: https://www.ntik.me/posts/voice-agent
Comments: https://news.ycombinator.com/item?id=47224295
By the way, if you go to https://github.com/claude and "block this user", every Github repo you visit containing code credited to Claude will actually have a warning sigil
CF
If you use AI-generated code, you currently cannot claim copyright on it in the US. If you fail to disclose/disclaim exactly which parts were not written by a human, you forfeit your copyright claim on *the entire codebase*.
This means copyright notices and even licenses folks are putting on their vibe-coded GitHub repos are unenforceable. The AI-generated code, and possibly the whole project, becomes public domain.
Source: https://www.congress.gov/crs_external_products/LSB/PDF/LSB10922/LSB10922.8.pdf
"As fun as it is to rib on Mozilla, the web needs Firefox. I feel for the Firefox developers who actually care. State of Mozilla will inspire no one. The sloppy prose are borderline unreadable. The presentation is designed to stop you reading."
I’m Tired And Everything Hurts
Jason Gorman
Hacker News
hanno
mcc
Jamie Gaskins
tante
Dave Rahardja
nixCraft 🐧