AI coding news — coding assistants, agents and benchmarks, from Cursor and Claude Code to Copilot and Lovable: the tools rewriting how software gets built.
Recursive released Flint, a free 40B open-weight coding model that runs on a laptop and was largely built by its Eureka Machine automated research system.
PHP2Shell: Wiz and OpenAI-linked researchers used AI to find two critical PHP flaws, CVSS 9.2, in 70% of websites. Patches shipped Oct 21.
LiteLLM 1.107.0 on PyPI carried a credential stealer and an hourly backdoor in all seven wheels, harvesting provider API keys before PyPI pulled the release.
Replit makes OpenAI its preferred partner and default for Replit Agent, CEO Amjad Masad says, while users can still choose Claude and other models today.
OpenCode says a distillation attack breach exposed OpenAI, Anthropic and Google API keys saved by 1.4% of Console users, who must revoke and replace them today.
Claude Code works from the terminal, Cursor rebuilds the editor, Copilot assists inside yours. We compare the three dominant ways developers code with AI.
Vibe coding means building software by describing it to AI and accepting what comes back. We explain the term, the workflow and why developers argue about it.
AI coding tools now span autocomplete, AI editors, terminal agents and app builders. We explain the four categories, the big names and how to choose well.