Today's AI landscape is defined by two converging forces: aggressive enterprise adoption and escalating geopolitical friction. Uber placed its entire 30,000-person workforce on ChatGPT Work, while Anthropic unveiled Claude Duet, its clearest bet yet on agentic computing. Meanwhile, DeepSeek reportedly raised $15 billion just days after new US restrictions, underscoring how capital continues to flow despite policy headwinds. On the infrastructure side, Meta is engineering a new stock class to fund its AI buildout, and a fresh analysis reveals the hidden storage costs of inference. Security concerns also loom large, with state-linked hackers exploiting enterprise software and critical PHP vulnerabilities rattling the open-source world.
Uber has rolled out ChatGPT Work to its entire global workforce of 30,000 employees, marking one of the largest enterprise-wide generative AI deployments to date. The move signals a shift from pilot programs to full-scale operational integration, with the company likely betting on productivity gains across engineering, operations, and customer support. The scale of the rollout also raises questions about data governance and how Uber will measure ROI on a tool now embedded in daily workflows.
Chinese AI lab DeepSeek has reportedly secured $15 billion in fresh funding just days after the US imposed new restrictions targeting advanced AI development. The timing is striking: rather than slowing the company, policy pressure appears to have accelerated domestic investment. The raise positions DeepSeek to continue scaling its models and infrastructure, and it underscores the widening divide between US-led AI governance and China's state-backed capital engine.
Anthropic has introduced Claude Duet, a standalone application built around agentic workflows rather than conversational chat. The app represents Anthropic's most direct bet yet that the next interface paradigm is agents that execute multi-step tasks, not chatbots that answer questions. Duet's launch puts Anthropic in closer competition with OpenAI's operator-style products and signals that the company sees agent orchestration as its core differentiator.
Meta is asking shareholders to approve a newly created class of stock designed to finance its sprawling AI infrastructure ambitions. The move is unusual and reflects the sheer capital intensity of training and serving frontier models at scale. By structuring the raise through equity rather than debt, Meta appears to be managing balance-sheet risk while keeping its buildout pace intact—a signal that the company expects AI capex to remain elevated for years.
A state-linked Chinese hacking group breached IT management giant ConnectWise and used that access to spy on thousands of downstream companies, according to a new report. The incident highlights the systemic risk posed by software supply-chain compromises: one vendor breach can cascade across entire industries. For AI companies increasingly reliant on third-party infrastructure, the episode is a stark reminder that model security means little without vendor security.
A new MIT Technology Review analysis examines the often-overlooked storage costs generated by AI inference—logs, embeddings, intermediate outputs, and cached context that accumulate at scale. While much attention goes to compute, storage is emerging as a quiet but significant line item in AI economics. For enterprises deploying models at scale, the piece argues, inference data retention policies will become as important as model selection.
Spotify is repositioning its AI efforts around agent compatibility—ensuring that when autonomous AI assistants recommend or play music, Spotify is the service they select. It's a forward-looking bet that discovery will increasingly be mediated by agents rather than human browsing. The strategy reflects a broader industry realization: as AI intermediaries grow, being the default integration point matters more than being the default app.
Walmart is actively rolling out the welcome mat for AI shopping agents, a sharp contrast to Amazon's reported efforts to block them. The divergence reveals a strategic split: Walmart appears willing to cede some interface control in exchange for broader distribution, while Amazon guards its storefront as a moat. How consumers and regulators respond could shape the economics of agentic commerce for years.
Recursive has released Flint, a laptop-friendly coding model that the company says was largely developed by AI itself. The claim—if verified—would mark a meaningful milestone in recursive self-improvement for narrow domains. Flint's local-first design also targets developers who want capable coding assistance without sending code to the cloud, a growing priority for privacy-conscious teams.
A new survey finds that while AI promises productivity, managers are largely experiencing friction—tool sprawl, inconsistent outputs, and unclear workflows. The gap between executive enthusiasm and frontline reality remains wide. The findings suggest that enterprise AI success depends less on model capability and more on integration, training, and process redesign.