OpenAI Raids Anthropic's Product Team and Ships GPT-Image-2.6 as All-Modality Race Heats Up
The product-aesthetics gap that made Claude Code the preferred tool for many developers is closing. OpenAI's design acquisitions and the arrival of Cat Wu, plus Google's rapid creative-model progress, mean that Codex and Gemini will soon match or beat Anthropic on experience, while adding video and multimodal workflows.
Cat Wu, the product lead behind Anthropic's Claude Code, has joined OpenAI, where she'll work on core products. OpenAI has also acquired NextSlides and Pierre, two teams known for strong product design, in an effort to shake off the reputation of being a pure engineering shop. The moves landed alongside the release of GPT-Image-2.6, which adds one-click meme generation, and the gray-scale beta of GPT-6.2-Astra, with leaked reports of significantly improved frontend and video capabilities.
Those releases fit the argument that the next phase of AI competition is all-modality and ecosystem, not just coding. GPT-6.2-Astra's video improvements have led to speculation that the dormant Sora video generator is being folded back into the product line, with GIF as the smallest sample of image-to-video. On the Google side, Gemini 4 Argon is said to perform well in creative tasks like writing, music, and video generation, with Carbon entering beta soon. A heavy Codex user reports that despite quota cuts and slowness, Codex remains the best value globally once resets and reset cards are counted.
Anthropic still leads in product experience and aesthetics, but only because it has been out front so long. The claim is that OpenAI and Google will catch up and surpass, and Anthropic's account-banning behavior is cited as a reason users are tired of it. For developers, the competitive pressure is shifting from raw model capability to product polish and multimodal integration.
The lack of non-compete restrictions for a core product lead moving directly to a competitor is rare, and it suggests either negotiated exits or a talent market where such clauses are unenforced.
OpenAI buying NextSlides and Pierre to fix its design reputation treats product taste as an acquirable asset, which is unusual for a company with OpenAI's engineering depth. It suggests the next AI differentiator is interaction design, not just model capability.
The framing of a GIF as the smallest sample of image-to-video implies video generation will not stay a standalone feature. If Sora's capabilities get wired into Codex, coding tools will start producing animated output as part of their normal workflow, which changes how developers demonstrate results.