via: https://news.ycombinator.com/item?id=47550282

What a breakdown, a baby philosopher, and a custodian taught me about the ceiling of single-agent AI

Practical techniques for steering GPT-5.4 toward polished, production-ready frontend designs.

Summary of the collected written experiences from developers of the rust compiler on using LLMs for programming and the deluge of LLM generated pull-requests.

A collection of skills to help LLM agents code in golang.

A storage that can be used to keep memory across sections.

This claims an innovative “holographic memory” that is apparently more compact and requires less infrastructure than RAG.

An in-depth article about working with LLMs from a senior engineering perspective.

An incredible article on the “anatomy” of large language models. Probably in large enough models some layers form a kind of circuitry for a specific purpose, e.g. solving math problems.

Convert PDF to markdown JSON quickly with high accuracy - datalab-to/marker

A tool that gives more context to agent, based on a curated set of API documentation.

via: https://news.ycombinator.com/item?id=47317132

Human DX optimizes for discoverability. Agent DX optimizes for predictability. What I learned building a CLI for agents first.

An LLM benchmark testing the model’s push back (or lack thereof) against BS.

MCP server that reduces Claude Code context consumption by 98%. 315 KB becomes 5.4 KB.