Anthropic presented dynamic workflows, a new multi-agent orchestration feature for Claude Managed Agents, showing how Claude can generate an orchestration program and coordinate sub-agents across phases of work. The feature was announced on Friday. Claude Console can now orchestrate longer, more complex multi-agent topologies without developers configuring coordinators and rosters for interconnections, though users can still supply specialist sub-agents and set session budgets.
The U.S. Air Force will purchase data from WindBorne Systems’ autonomous weather balloon fleet under a new contract that could reach nearly $50 million for the startup. The contract is worth up to $49 million, with $1.7 million already ordered so far and further orders expected. This increases WindBorne’s funding for scaling its 700-balloon constellation and supports additional research and development on balloon hardware and performance.
Ray Serve LLM Deep Learning Containers were used on Amazon EKS to deploy a single large language model split across two GPU nodes, with one node handling prefill and the other handling decode behind one OpenAI-compatible endpoint.
Microsoft will scale its in-house Maia inference chips across Azure starting in 2027 to reduce reliance on Nvidia. The Maia 300 is scheduled for first production deployments in 2027, with Microsoft seeking capacity for more than 300,000 chips for that year. Microsoft is shifting AI inference for its own products onto Maia 300 hardware and expanding Maia use for model serving while waiting for independent benchmarks.
Raven, a Graylark Technologies AI tool, was purchased by local US law enforcement to geolocate specific homes using interior photos drawn from listings like Airbnb and Zillow. The demo shows results in about 2 seconds after uploading a photo. As a result, police can match a single interior image to a particular property and then obtain a narrowed set of likely listing identities.
Microsoft announced its third-generation Maia 300 in-house AI inference chip as the primary inference system for its products and services. It targets 40% more tokens per megawatt than leading GPU-based systems. Maia 300 will begin rolling out to production in 2027, reducing Microsoft’s reliance on Nvidia GPUs for inference workloads.
Microsoft launched Maia 300, its next-generation AI inference platform for long-context reasoning, low-latency conversation, and agents on Azure. It is built around 8 HBM4 stacks and 96-accelerator racks, with a claim of 40% more tokens per megawatt. Production is set to start in 2027 and Maia 300 is intended to become the main inference system for Microsoft’s products as Azure scales to multi-gigawatt deployments.
llm-d’s distributed inference framework coordinates clusters of model servers so large LLM inference can handle many concurrent requests, reuse shared prompt context, and manage overload. In a demo, prefix-aware routing captured 97% of the available cache reuse in the tested workload. As a result, redundant prefill work drops and traffic is paced and routed more efficiently across servers, including when requests arrive in bursts from AI agents.
Every AI story that matters,
in your inbox by 8am.
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the
day in two minutes. Follow companies and topics for alerts, or get the
briefing in Slack. Free, no spam, unsubscribe anytime.
Reading TLDRocket needs no cookies, and the readership counts we rely on come from
our own cookieless analytics. Google Analytics is the exception: it sets cookies and
reports to Google, so it stays switched off until you allow it. You can change your
mind any time from “Cookie settings” in the footer.
Strictly necessary
Session security and form protection (tldrocket-session,
XSRF-TOKEN, 2 hours). The site cannot work without them,
so they need no consent.
Always on
Google Analytics 4 (_ga,
_ga_<id>, up to 2 years). Measures which
stories and sections readers use. Google acts as a third-party processor and may
store the data outside the EU. No advertising, no profiling, no data sold.
