AI digest: shifting sands
OpenAI sheds side projects whilst Anthropic expands into design, plus major funding moves and enterprise tooling updates.
Another week of big moves and strategic pivots. The industry’s settling into clearer patterns around what actually works versus what sounds clever.
OpenAI cuts the fat, loses key talent
Kevin Weil and Bill Peebles have left OpenAI as the company shuts down Sora and folds its science team. This signals a sharp pivot away from consumer moonshots toward enterprise AI. Makes sense given the revenue pressure, but losing the head of product and a key research leader suggests the transition isn’t going smoothly.
Cursor chases the $50bn club
The AI coding editor is reportedly raising over $2bn at a $50bn valuation, driven by enterprise adoption. That’s GitHub Copilot money for what’s essentially a souped-up VS Code fork. The bet here is that developer tooling becomes the new operating system, which isn’t mad given how much code AI is writing these days.
Anthropic goes visual with Claude Design
Claude Design lets users create prototypes, slides, and marketing assets through conversation. Smart move targeting the non-designer market that Canva dominates. The real test is whether Claude can handle the iterative back-and-forth that design work requires without driving users mental.
Google’s Auto-Diagnose tackles test hell
Google released Auto-Diagnose, an LLM system that reads integration test failure logs and identifies bugs automatically. Anyone who’s stared at thousands of lines of test output knows this pain. If it actually works at scale, this could save serious developer time and sanity.
Red teaming goes mainstream
A comprehensive roundup of 19 AI red teaming tools highlights how security testing has moved from niche practice to regulatory requirement. Tools like Mindgard, Garak, and Microsoft’s PyRIT are becoming essential as AI systems handle more sensitive tasks. The regulatory pressure is real and these tools are catching up fast.