Tag: analysis

All the articles with the tag "analysis".

• Slop Is Moving Up the Stack

  Published: 12 Nov, 2026 |  at  10:15 AM

  [code]

  Agents write cleaner lines of code than a year ago. The slop moved into design: hand-written copies of libraries, catch-all error handling, and libraries other agents don't use.

• Can Agents Design Libraries for Agents?

  G. Orlanski, A. Zhang, A. Trost, V. Chen, F. Sala, A. Albarghouthi, and L. Schmidt

  View code on GitHub Paper View paper on arXiv | 2026 | Technical Report

  Introducing LibraryDesignBench, a new benchmark that scores agent-written libraries solely by how much they help the future agents that use them.

• GPT-5.4 Writes Clean Code That Fails More Tests

  Published: 6 Mar, 2026 |  at  06:00 AM

  [code]

  GPT-5.4 writes the cleanest code of any GPT model we've tested — and fails more tests than 5.3. Erosion drops, duplication drops, but pass rate and core both regress. We dig into why.

• Opus 4.6 and GPT-5.3 Codex Score Higher, but the Code Is Still a Mess.

  Published: 11 Feb, 2026 |  at  06:00 AM

  [code]

  Anthropic's Opus 4.6 copy-pastes. OpenAI's GPT-5.3 Codex over-abstracts. Both miss edge cases at the same rate. New SCBench results with a guide for when to trust each model.