DeepSeek DeepInfer 🐋, PyTorch on TPUs ⚡, Ramp’s agent framework 👨‍💻. TLDR AI 10/21

The biggest model. Maximum reasoning. Let it run. AI coding costs can climb fast. Simply capping usage can get in the way of productive work. Databricks found much bigger levers: model choice, routing, reasoning effort, context management and smarter guardrails. See what we learned managing AI coding at scale. Keep teams moving without giving AI spend a blank check.

See how to cut AI coding costs

See AI spend controls in action

Watch the webinar: Govern AI Spend at Scale

Voxtral 3 offers advanced speech-to-text and text-to-speech capabilities across 50+ languages. Featuring models like Small Transcribe 3 for multilingual transcription and Mini TTS 3 for natural emotional speech, it includes a voice agent SDK combining real-time listening, reasoning, and speaking, enabling seamless interactions and task execution over phone or web.

TorchTPU is an open-source PyTorch backend designed to make running and migrating models to Google's TPUs easier. It supports training and inference through optimized eager execution, torch.compile integration, and native hardware acceleration.

In less than 10mins, you can build and deploy an AI voice agent with LiveKit for free. 1M people have already built with LiveKit and the platform powers AI voice agents at Spotify, Headspace, and xAI.

Try it free → cloud.livekit.io

llm-d is a Kubernetes-native stack for distributed LLM inference. It runs vLLM, SGLang, and more across clusters, turning single-node engines into production-grade serving on local hardware. llm-d provides tested deployment recipes for common production patterns that are adaptable starting points across models, hardware, and workloads. It runs production LLM inference across GPUs, TPUs, XPUs, CPUs, and emerging NPUs with consistent performance patterns, so teams can deploy once and scale anywhere with predictable cost and behavior.

DeepInfer combines distributed inference, caching, load balancing, and speculative decoding to improve efficiency. DeepSeek processes 155 trillion tokens daily per cluster and earns $38 in revenue per GPU-hour while open-sourcing much of its inference framework.

Vercel's durable-harness allows agent harnesses to run durably in the cloud out-of-the-box. It allows users to run Pi, OpenCode, Codex, and Claude Code durably on Vercel or self-hosted on their own Postgres instance. durable-harness ensures that sessions survive crashes mid-turn and keep running even if the browser disconnects. Turns that lose function partway through continue from the log on a new one without manual retry and without repeating what already finished.

Laminar is an in-house agent harness for Ramp's agentic products. It features secure conversation history storage, session compaction, standard interfaces for tool calling, message queuing and steering, subagents, multiplayer chat support, observability tooling, and much more. The first version of Laminar was built in six weeks with a small team of obsessive, experienced engineers who worked with and steered the agents. Its code was generated entirely by agents using the most powerful models available at the time. While building in-house has become cheaper, Ramp still needs good engineers to be on-call, keep the platform healthy, and figure out what product teams actually need and how it fits into the bigger picture.

Google CEO Sundar Pichai discussed the company's commitment to frontier AI research, including sacrificing short-term cloud revenue to allocate compute to model development.

TLDR is hiring a senior engineer to build the agent-first operating layer used across the company. We're looking for a builder who has taken an LLM agent to production and still owns it. Click here to learn more.

OpenAI acquired Orchestra Bio to combine its AI-powered drug development coordination platform with Rosalind, its suite of life sciences AI tools.

On September 8, Jacob Coxon, a 27-year-old machine learning researcher who had resigned from Anthropic, warned the world that AI companies were not acting responsibly and that they were gambling with people's lives by racing straight to self-improving superintelligence. Many researchers seriously believe that AI could kill all humans. While Anthropic may be trying its best, it still doesn't have a plan to solve alignment for superintelligence and is not on track to.

Learn how to optimize and scale AI workloads at the Akamai AI Summit at Oracle Park. Discover new ways to maximize model performance and solve infrastructure challenges.

🔗 Register for free

See how a gateway governs what agents access and what they spend plus, Anaqua's CIO on their rollout.

OpenAI introduces two metrics to address limitations of g-mean² in measuring model monitorability.

The interface between the digital and physical world is lossy - this lossiness will shape when and how future superhuman hardware engineers will emerge.

Eleven Realtime is a new class of Speech-to-Speech models for building natural voice agents.

Meta's AI companion Muse reached 10.2 million US downloads within 40 days, a milestone previously achieved by only 11 other apps.

MemCurrent is a persistent memory architecture for production agents that can triple an agent's pass rate over no memory while also making it faster and cheaper.

Locked chats are kept separate from regular chat history, pinned conversations, and chat search results.

SparkLLM released Spark-X2.6-4B, an open-source model with improved coding, reasoning, long-context, and agentic capabilities through post-training.