Articles from lucumr.pocoo.org

The author tests Pangram’s AI detector on LLM‑generated and manually rewritten text, finding both still flagged as AI, highlighting limits of current detection methods.

Armin Ronacher argues that Astra’s token‑efficient coding yields unreadable slop, wasting effort and challenging its usefulness for real software engineering.

LLMs diminish the need to learn language specifics, letting developers choose languages like Rust and Zig for speed and size, and tackle complex low‑level projects without deep expertise.

AI-assisted coding removes collaborative friction, letting individuals make changes without shared understanding, causing codebases to grow incoherently like a never‑ending Tower of Babel.

The article argues that while 'harness loops' atop coding agents accelerate porting, experimentation, and security, they degrade code quality and comprehension, creating machine-dependent codebases.

Armin Ronacher argues for the term "clanker" as a deliberately mechanical, dehumanizing label for LLM-based tools to resist anthropomorphism and keep responsibility clearly on humans.

LLM-generated slop in issue trackers and code contributions harms open-source maintenance, forcing maintainers to resist local workarounds that undermine global invariants.

Armin Ronacher argues that as GitHub declines due to instability and product churn, open source must preserve its history by moving to decentralized, archivally-funded homes that avoid the pre-GitHub era's fragility.

Pi, a minimal coding agent with a tiny core and extension system, shows how software can be built by agents for agents, pointing toward the future of development.

Armin Ronacher reflects on 2025 as a year of agentic coding tools like Claude Code, which replaced much of his direct programming, and discusses the cultural, technical, and social challenges this shift presents for software engineering.

Claude Code's plan mode is a thin prompt overlay and read-only guard, not a fundamentally different execution path.