From the archive
Thursday, 19 November
Folded from 36 sources that day.
Sorted by quality and by how many sources covered it, the same for everyone.
-
An engineer didn’t message his AI for 22 of 44 days. It built an Android app anyway.
Cognition launched Devin Bots, persistent agents that oversee multiple Devin coding sessions. It used one to build an Android app, merging 1,134 pull requests over 44 days.
-
How OpenAI’s Performance Factory keeps ChatGPT fast
OpenAI describes an autonomous Codex-based system that investigates performance regressions and proposes fixes. In three months it tracked over 1,700 investigations and root-caused more than 600 cases.
-
Gemini 4 Neon: GDM's Opus/Sol Contender
Google's midtier Gemini 4 Neon costs $3 per million input tokens and $15 output, with a 1M-token context. It places second on the Vals Index and Agent Arena, and Cursor and GitHub Copilot support it on day one.
-
Google’s Gemini 4 Neon tripled its coding score without changing the model
On the SWE-sweep benchmark, Gemini 4 Neon rose from 12.8% under mini-SWE to 39.0% inside Antigravity, showing how much the coding harness affects results.
-
Design for Understandability
A Google talk argues that as LLMs generate more code, developers' key contribution is judging whether code is healthy. It proposes design by contract, class invariants, and modularity.