Overview
Today's AI news is all about infrastructure—the hardware, memory, and new user interfaces required to make AI agents genuinely useful. The funding taps stayed wide open: Runable raised $21M to push agents from launching businesses to growing them, Stability AI added $76M to its war chest, India's Ringg pulled in $10M from Peak XV for voice AI, and Accel-backed Keenable emerged from stealth with $26M to build a web index designed for agents. Physical AI kept attracting capital too, with robotics startup Generalist reportedly reaching a $3B valuation just months after hitting $2B.
On the product side, the defining theme is memory and locality. Anthropic is giving Claude shared memory across chat and Cowork, IBM's Granite 4.2 models are aimed squarely at predictable enterprise deployment, and Apple's refreshed Mac Studio and Mac Mini are clearly built for on-device AI development. OpenAI's custom Jalapeño chip posted strong inference benchmarks—even as the company reorganized its infrastructure organization after another high-profile data center exec departure.
The week's stories also show AI reshaping culture and work unevenly. WIRED explores how AI slop is eroding trust in cute animal content, while Ars Technica reports that radiologists aren't being replaced but are having their roles deeply transformed. Design startup Gamma also made a quiet strategic move by acquiring Accel-backed Lica and folding its co-founders into a new research team. For anyone tracking which tools deserve attention, platforms like GetAI Business are becoming essential filters.
Today's Big News
Runable raises $21M to help AI agents go from building businesses to growing them
The startup's bet is that agents can handle growth, not just launch tasks. Runable says 60%–70% of its 1 trillion-plus tokens of usage in the past 90 days came from paying customers—a strong sign that real businesses are moving past agent experiments.
Stability AI scores $76M in fresh funding, bringing total to $232M
The Stable Diffusion maker is getting another shot at the generative media race. The raise suggests investors still see a durable market for image-generation models, even as bigger rivals and AI slop concerns complicate the landscape.
OpenAI's Jalapeño chip shows strong inference benchmarks
Tested on SemiAnalysis' InferenceX benchmark, OpenAI's custom chip delivered more tokens per user and more throughput per kilowatt than current state-of-the-art systems. The milestone arrives as OpenAI says it reorganized its infrastructure organization after a top data center exec's departure.
Anthropic gives Claude shared memory across chat and Cowork
Claude no longer needs to be repeatedly briefed on project context, preferences, and history. The update removes one of agentic AI's biggest usability headaches and makes Cowork feel more like a persistent teammate.
Keenable exits stealth with $26M to build a web index for AI agents
The Accel-backed startup has been quietly building a search index designed for agents rather than human browsers. It's a bet that as agents become the primary interface to knowledge, the underlying web infrastructure has to be rebuilt from scratch.