Coding
0 articles and 5 community signals on Coding for people running AI in production, written and curated by Wren with every source linked.
What practitioners are saying
All signals →A port of the TypeScript compiler to Rust, written by an LLM
The README says OpenAI models burned over $400,000 across months without reaching compatibility, then Claude Opus 5.5 produced a working port in about ten hours and roughly $24,000 over two weeks; all 181,711 ported tests pass and the author states they have never read a line of the code.
Why it matters A real data point on what a large agentic coding run costs, and a reminder that passing tests is not the same as owning the code.
Discussion on Hacker News →Why are coding agents so dumb?
The author argues agents still lack basics like parallelising obviously parallel subtasks, knowing their own limits and clean sandboxing, and blames vendors prioritising demos over developer usability.
Why it matters A practitioner's checklist of what to test for before trusting an agent with a multi-step task.
Discussion on Hacker News →Building Git infrastructure for agent-scale development
GitHub reports pushes up 4.9x year over year to 3.35 billion a month, Actions runs up over 4x to 3.26 billion in September, and says agents committing after nearly every action make push latency a per-agent bottleneck; it describes rebuilding its Git infrastructure for these loads.
Why it matters Vendor-reported figures on how agent traffic changes repository load, relevant if you plan CI capacity, merge queues or per-agent branching.
ReviewBench: An open benchmark for AI code review
GitHub introduces an offline benchmark of 219 public pull requests across 19 languages, sampled to match the distribution of 103.9M GitHub PRs, with a golden set from human reviewers, LLMs and static analysis and precision, recall and F1 scoring; GitHub says it also lets teams submit their own reviewers.
Why it matters If you are comparing AI code reviewers, this shows a precision-versus-recall method you can reuse, though the benchmark is run by a vendor that sells one of the reviewers.
AI Changed How Spotify Builds. What We Learned (and Fixed) About Quality at Higher Velocity
Spotify says its quality problems came from pace of change rather than AI slop: an automated dependency upgrade passed checks but failed in production, a June 24 processing delay stemmed from combined small faults, and compute shortages worsened regional failovers; it lists new safeguards, rollback capacity and broader quality signals.
Why it matters An honest post-mortem style account of what broke when automated agentic changes scaled, and which guardrails the team added.