Here’s what’s worth your time today.
The NSA is reportedly using Anthropic’s unreleased Mythos model for cybersecurity tasks, despite the Pentagon labeling the company a national security risk after a dispute over military use. Mythos was deemed too powerful to release publicly and is limited to a small group of organizations, but intelligence agencies appear to be among them. The situation highlights the reality of AI right now: even when governments push back publicly, they’re still quietly relying on the same frontier systems behind the scenes.
Supported by Bland AI
Norm lets you one-shot prompt a fully structured voice agent. Just describe what you want—appointment scheduling, customer support, lead qualification—and watch Norm build the agent, pathways, and logic for you. Safe branching. Agent-on-agent testing. Deploy when you’re ready.
Google is doubling down on custom AI chips, with new TPUs designed specifically for inference as the battleground shifts from training models to actually running them at scale. Demand is already surging, with deals signed with players like Meta and Anthropic, and Google now has a key edge few others can match: it builds both the models and the hardware. The bigger signal is where the AI race is heading next. It’s no longer just about who has the best model, but who can deliver results faster and cheaper, and that puts chip design at the center of the fight.
Marc Benioff is pushing back on the idea that AI agents will wipe out SaaS, arguing instead that they make platforms like Salesforce more valuable, not less. While investors worry about shrinking seat-based revenue as AI does more work, Salesforce is leaning into agents with new products like Agent Albert and shifting toward usage-based pricing. Early traction is mixed, but customers are already using agents to cut support and IT workloads by up to 40%. The bigger signal is that incumbents aren’t being replaced yet — they’re adapting, turning AI from a threat into a layer that deepens their grip on enterprise workflows.
Around half of UK executives now expect AI to reduce overall employment, up sharply from just a few years ago, with entry-level roles seen as the most exposed. Demand for junior positions is already being rethought, while daily AI usage among workers continues to rise quickly. The gap is in execution: employees are using AI to improve output, but companies haven’t yet redesigned workflows to fully capture the gains. The bigger signal is a shift in mindset, where leaders are increasingly viewing AI as a cost-cutting tool, even as the data suggests the real upside may come from revenue growth instead.
Meta is planning to lay off around 10% of its workforce in May, with more cuts expected later this year as it restructures around AI. The move follows a broader shift across tech, where companies are investing heavily in AI while reducing headcount tied to traditional roles and management layers. Meta is simultaneously building new AI-focused teams and pushing toward more autonomous systems that can handle complex work. The bigger signal is becoming hard to ignore: AI isn’t just augmenting jobs anymore, it’s starting to replace parts of the org chart.
Cerebras’ IPO filing shows how tangled the AI ecosystem has become, with OpenAI not just as a customer but a major stakeholder tied to massive compute commitments. The chipmaker is targeting a $35 billion valuation, with ambitions up to $250 billion, backed by deals that blur the line between financing and infrastructure access. The bigger signal is that AI companies aren’t just raising capital anymore — they’re locking in long-term partnerships where compute, equity, and distribution are all part of the same deal.
<img src='https://www.vpdae.com/open/8a68aa83.gif?opens=1' width='1' height='1'>