Urgent.News

What's breaking now, across thousands of outlets.

AI

I gave Gemini 4 Argon in Antigravity and Claude Code the same messy side project, and only one of them left it better than it found it

Vibe coding only goes so far

I gave Gemini 4 Argon in Antigravity and Claude Code the same messy side project, and only one of them left it better than it found it

Goldman Sachs has restricted the use of AI coding tools, such as Anthropic’s Claude Code and Microsoft’s GitHub, following a review regarding risk and productivity. According to the Times of India, access is now limited to engineers and strats, while most non-technical staff, including bankers and traders, are banned from using them. This decision has caused frustration among traders who utilized the tools for client trading flows and financial modelling.

In a separate test reported by XDA Developers, Google's Gemini 4 Argon in Antigravity and Claude Code were both given the same messy side project repository and a one-line brief to complete within an hour. The test utilized a codebase from a 2015 Amazon Kindle Paperwhite that had been repurposed into a Home Assistant dashboard.

Brief written by urgent.news from XDA Developers, Android Police, Times of India — 3 reports on this story. Machine-written — may contain errors; check the original before relying on it.

This story

This is one outlet's version. Read the fullest account.

Read the original at xda-developers.com →

More in AI

Come Code, IDE e Wiki di osFoundry rendono revisionabile il lavoro di più agenti AI in parallelo

Sono le 16 e da dopo pranzo tre agenti di coding lavorano sullo stesso repository. Uno ha sistemato l'arrotondamento del totale del carrello, uno ha aggiunto un codice sconto, uno ha scritto i test.

  • osFoundry unifica chat, agenti AI, file e strumenti in una singola finestra
  • Sistema consente assegnazione di task a più agenti AI contemporaneamente
  • Funzionalità di revisione facilitano la collaborazione e riducono conflitti

LLM Returns Invalid JSON Once in a While: Structured Outputs vs Validate-and-Retry

TL;DR: If json.loads() on model output fails only occasionally, there are three distinct causes: truncation, stray prose or code fences around the JSON, and schema drift (valid JSON, wrong shape).

  • LLM JSON parsing failures can stem from truncation, extraneous text, or schema mismatches.
  • Structured outputs and typed models help eliminate extraneous text and schema mismatches.
  • Validation-and-retry approach with logging raw responses is recommended for handling failures.

More from Friday 6 November →