AI moves from model launches to the messy work of making it useful

Salesforce’s reported $5.7 billion acquisition of customer-engagement platform Braze has drawn criticism from competitors. For teams choosing marketing tools, it’s a reminder to weigh product fit and flexibility alongside the growing pull of platform consolidation.

A report says Gemini 3.6 Flash in Antigravity can use up to 17% fewer tokens. If that holds for your workloads, it could ease usage costs—but it’s worth testing on your own tasks before changing a production setup.

Maxinsights has raised $30 million in seed funding and introduced two tools for collecting and processing real-world training data. The announcement highlights a practical bottleneck for robotics and other physical-AI projects: getting useful data from the world into a form models can learn from.

UnternehmerTUM, Microsoft, Infineon and NVIDIA have launched an equity-free accelerator focused on physical AI. For early-stage teams, access to structured support may help turn prototypes into products without giving up equity at the outset.

A report points to a steady run of new model launches from OpenAI, Google, xAI and Anthropic since September 12, despite Dario Amodei’s call for labs to pace releases. Builders may have to balance the appeal of new capabilities against the costs of repeatedly evaluating and adopting them.

Japan’s AI&Co has launched a video tool that lets enterprises build dedicated models using characters and video assets they have rights to. That rights-first approach could matter to studios and brands looking to use generative video while keeping control of the material behind it.

A new French literary prize will recognize fiction written with AI assistance, with authors expected to disclose the tools they used. Clear disclosure rules could give writers and readers a shared starting point for discussing what authorship means in AI-assisted work.

NVIDIA’s GTC Berlin brings developers, researchers and industry leaders together to discuss what’s next in AI. For people building in the region, the event is a chance to follow how European talent and organizations are shaping the wider conversation.

JFrog has introduced local checks intended to flag unapproved AI tools on developers’ laptops before they reach central security controls. That could give security teams earlier visibility, while raising the familiar question of how to add oversight without disrupting developers’ workflows.

Martingale has released Raven Alpha, a model-drift detection method, according to the report. For teams running models over time, drift monitoring is part of catching when real-world behavior starts to diverge from what their evaluations led them to expect.