AI digest: Big moves and reality checks
Major model releases from Anthropic and xAI, while research shows AI dependency might be happening faster than we thought.
The week brought proper updates from the big players, but also some sobering research on how quickly we’re getting dependent on AI assistance.
Anthropic pushes Claude Opus 4.7 for serious work
Anthropic released Claude Opus 4.7 with major improvements for agentic coding, high-resolution vision, and autonomous tasks. This feels like the first real push towards models that can handle proper software engineering workflows rather than just code suggestions. The timing is interesting given all the focus on AI agents actually shipping products rather than just demos.
xAI enters the voice API battle
xAI launched standalone speech APIs built on the same infrastructure powering Grok Voice across Tesla and Starlink. Smart move to monetise existing tech, though the voice API market is already pretty crowded. The enterprise focus suggests they’re going after the boring but profitable business use cases rather than consumer flashiness.
Ten minutes of AI use measurably hurts problem-solving
New research found that just 10-15 minutes with an AI assistant weakens problem-solving ability and persistence. This is genuinely concerning if true. We’re all using these tools daily now, and if they’re making us worse at thinking through problems independently, that’s a proper issue worth tracking.
Meta reportedly trading people for compute
Meta is preparing to cut around 8,000 jobs to fund AI infrastructure spending, with potentially 20% of the workforce affected overall. The “headcount for compute” trade-off is becoming the new normal across big tech. Brutal but probably necessary given the capital requirements for frontier AI.