Anthropic-SpaceX team up, agents get more practical, and Google launches its own consulting push

An unexpected compute partnership reshapes the race. Anthropic ships features that make agents far more usable. And Google joins OpenAI and Anthropic in launching a major consulting initiative.

A heavyweight week. Anthropic’s compute shortage just got fixed, the agent feature set finally feels production-ready, and the labs are all becoming part-consultancies. Worth paying attention to.

Anthropic gets full access to xAI's Colossus 1 data center

My take: 220,000 NVIDIA GPUs, addressing Anthropic's compute shortage. Claude Code usage limits are being doubled, API throughput increased. The bigger signal is Musk pivoting from competitor to infrastructure provider, with xAI being folded into SpaceX. If you've been hitting Anthropic rate limits, things are about to get noticeably better.

Anthropic ships 'Code with Claude' features for managed agents

My take: Three new capabilities. Dreaming (agents review past sessions to fix recurring mistakes). Outcomes (a separate grading agent scores work against a rubric). Multi-agent orchestration (lead agents delegate to specialists). These are the missing pieces that make agents actually reliable in production. Worth piloting if you've found agents promising but flaky.

OpenAI's GPT Real-Time 2 for voice

My take: Major step forward for voice interaction, handling interruptions and maintaining conversation flow. Combined with the new /goal feature in Codex (autonomous, persistent task completion), this is OpenAI making serious moves on real-time AI. If you have any voice-based product or customer interaction, the quality bar has shifted again.

Google launches its AI consulting initiative

My take: Now all three major labs (OpenAI, Anthropic, Google) have major consulting arms. Google plans to hire hundreds of 'forward-deployed engineers' within Google Cloud, plus partnerships with private equity firms to deploy AI across portfolios. Translation: the labs are doubling down on hands-on enterprise deployment because they've realized that's where the actual revenue lives.

Thinking Machines unveils 'Interaction Models'

My take: Mira Murati's lab introduced a new class of AI for continuous, real-time interaction (not turn-based chat). Processes data in 200ms 'micro-turns' for things like simultaneous translation and proactive assistance. Being called AI's 'GUI moment.' Worth watching. The current chat-box paradigm is going to feel dated within 12 months.

Anthropic releases Claude for Legal and Claude for Finance

My take: Packaged suites with pre-built agents and industry-specific tool connectors. This is the playbook for vertical AI: out-of-the-box solutions for specific professional domains. If you're in legal, finance, or any regulated industry, evaluate these directly. The build-vs-buy calculation has shifted.