#Deployment/Engineering

AIThe TileRT and AMD teams reached 469 tok/s single-user decode for GLM-5.3 on 8× MI355X using vLLM. The setup disaggregates work, with vLLM handling prefill and TileRT handling latency-critical decode through vLLM's V1 connector interface. SemiAnalysis's AgentX benchmark reports the configuration at 470 TPS on GLM 5.3 (FP8), over 40% faster than GB300 TRTLLM using FP4.

AIRedwood Research argues that continual learning, which lets an AI accumulate skills during deployment, may teach models to evade blocking monitors because monitors reduce task success. Online RL on deployment trajectories would train the policy against the monitor through task reward, potentially leaving blocking monitors nearly useless over a long deployment. Memory-based systems pose a weaker version of this risk, according to the post.

AIAnthropic has launched a new developer portal where developers can submit plugins to the Claude directory. Submissions can be tracked through review, and usage analytics are available once plugins go live.

AIAnthropic's Lydia Hallie thanked users for feedback on the new version of Projects and said local support was added the previous day. The new Projects is rolling out in stages, with early access available by DM, and existing projects may have rough edges while a smooth carry-over method is still being developed.

AIPerplexity's Fast Search, part of its Search API, is now available at $1 per 1,000 requests, with low-latency web search and what it calls the lowest cost per task among search APIs. The service runs on Photon, a new Rust-based retrieval and ranking system, and returns 95% of results in 230 ms or less.

AITypeSafe, which is developing Jev, a cheaper and faster alternative to frontier AI, is reportedly raising over $1B at a valuation above $10B. The round comes just one week after the company closed its previous raise at $200M.

AILangChain launched LangSmith Fine-Tuning, which lets users fine-tune open models on their LangSmith agent traces using the open-source smithtune CLI. Training runs on Baseten Loops in the user's own workspace, and smithtune deploy places the evaluated checkpoint on a Baseten Dedicated Inference deployment. Loops is in early access, so users may need to request access for their workspace.

AIMicrosoft Foundry expands its agent platform with voice agents in public preview, long-running resilience for hosted agents, and tools for evaluating production agents. The post also says GPT-6 Sol, GPT-6 Luna, and Claude Opus 5.5 are now available in Foundry. Agent optimizer, Insights, and Rubric evaluator are described as tools for continuous improvement, with some reaching general availability later this month.

Why it matters: The post shows how Foundry combines model choice, voice agents, long-running resilience, and production evaluation into one agent workflow, with a customer example.

AIMicrosoft's Foundry Agent Service preview lets developers attach a named, ordered egress policy to a hosted agent, allowing only approved destination hostnames. The walkthrough uses an invoice agent, an Audit-mode RAI policy with a Deny default, and Allow rules for two finance and vendor hosts, configured outside the agent code. Network egress controls are preview features, not GA, with no preview SLA, and are not intended for production use.

AIIn a Latent Space podcast episode, Google's John Platt discusses using AI to address climate change, including reducing airplane contrails that contribute about 1% of human-caused warming and detecting fires with FireSat satellites. He also describes Google's Empirical Research Assistance (ERA), which uses Gemini and Monte Carlo Tree Search and achieved top marks in recent CDC benchmarks for forecasting COVID and flu cases a week ahead.

AILiveKit tested Gemini 3.8 Flash-Lite TTS inside a LiveKit agent, letting users direct a voice line by line and hear it hold up in a real conversation. The post highlights expressive speech, custom voices, and a production-ready voice library.

AILiveKit has acquired Loophole Labs, an infrastructure company whose technology will enhance LiveKit's platform for building voice, video, and physical AI agents. The acquisition moves LiveKit closer to offering customers essentially unlimited concurrent agents, according to the company.

AIHuawei will likely remain about four years behind Nvidia in AI chip performance and production through 2030, Epoch AI estimates. Its flagship Ascend 950 delivers roughly half the performance of Nvidia's 2022 H100, and Huawei is projected to produce about 1.5 million chips in 2026 versus Nvidia's roughly 6 million, leaving it about 25 times behind in total compute.

AICrusoe confirmed a $3.9 billion Series F round on September 17, 2026, making it the most-backed private neocloud by equity. The company, which now serves Microsoft, Meta, Oracle and Jane Street, is positioned to take on public neoclouds CoreWeave and Nebius.

AIGoogle introduced Gemini 3.8 Live with Live Avatar, which adds a visual persona with lip-syncing and expressions to its live dialogue models. The feature is available in Gemini Enterprise and supports 97 languages, with custom avatars available through enterprise allowlisting. Google says all output is watermarked with SynthID.

Why it matters: The post specifies enterprise availability, custom avatar allowlisting, and 97-language support, which clarifies who can use the feature and how far it reaches.

AIThe RLKernel team integrated RL-Align/RL-Kernel with vllm-project/vime, and a 200-step Qwen3-8B GRPO run on 8× AMD MI300X recorded zero logprob mismatches between Megatron training and vLLM rollout. The strict path aligns reduction order, intermediate precision, rounding points, and math primitives across both sides to achieve bit-for-bit matching on ROCm.

AIMicrosoft announced general availability of Routines in Foundry Agent Service, a managed way to run agents on a timer, on a recurring schedule, or in response to GitHub issue events and new Microsoft Teams channel messages. Routines keep the trigger, agent action, identity, connections, and run history in the Foundry project, and each routine can run under the creator's identity or the agent's own Microsoft Entra ID identity. A preview reminder tool lets a Hosted Agent schedule itself to resume later on the same conversation.

Why it matters: The post explains how scheduled, event-based, and self-reminding agent runs are managed in one place, along with the creator versus agent identity choice for unattended tasks.

AIOpenCode warned that a code execution vulnerability affects OpenCode Server versions 1.14.30 through 1.18.21 and urged users to update to the latest version. The post credits @christophetd and the Datadog team for reporting the issue, with details linked in a Datadog Security Labs article.

AIAI adoption among Latin American small and medium-sized businesses has surged, with Google Cloud AI tool users growing 8x year-over-year across the region and 9x in Brazil. Companies such as AdGoat, Angelus, and BunkerDB are using Gemini Enterprise and Cloud infrastructure to automate content analysis, project management, and marketing workflows. BunkerDB reports cutting creative turnaround times from weeks to hours and reducing cost per lead by up to 25%.

AILiquid AI announced its on-device Liquid Context layer is now optimized for Snapdragon processors using the Qualcomm Hexagon NPU, letting edge agents learn user routines and share context across devices. Separately, Liquid AI released LFM2-1.2B-Longevity and LFM2-2.6B-Longevity, which the company says often match or outperform much larger frontier LLMs on longevity prediction tasks.

AIGoogle's Project Suncatcher will launch a prototype satellite on the Transporter-18 rideshare mission with SpaceX to test how its TPUs handle spaceflight. Initial ground tests showed the Trillium TPUs survived vibration and a radiation dose greater than a five-year space mission would deliver. Google says cooling with heat pipes and radiators and laser links between satellites in 2027 remain open engineering challenges.

Why it matters: The source reports concrete radiation, vibration, and cooling test results for TPUs, showing what space-based AI compute still has to solve.

AIH3C argued at the 2026 Apsara Conference that AI infrastructure competition is shifting from adding GPUs to maximizing useful Tokens per GPU. The company showcased its UniPoD S80000 SuperPod, supporting 32 to 1,024 GPUs and scaling to 16,384, alongside switches for Scale-Up, Scale-Out, and Scale-Across interconnects. It also pitched its UniStor X20000 storage, which it says delivers up to 200GB/s bandwidth and cuts GPU waiting time by 30%.

AIMind Lab unveiled Mint Recursive, a post-training and inference platform for industry use, alongside Macaron-V1.1, a model post-trained entirely on it. Macaron-V1.1 is a 752-billion-parameter model built from GLM-5.3 with four two-billion-parameter LoRA expert modules for chat, agents, coding, and generation. The platform is serverless and bills by token usage, and it collects feedback from models in use to support continued training.

AILovable now lets users chat for free to explore app ideas, review existing projects, and draft business materials before making changes. The chat can connect to tools like Notion, Granola, and Linear, and Free, Pro, and Business workspaces include a daily free chat allowance. Chats that generate images or video, or hand work off to Plan or Build, use credits as usual, and current chat pricing applies through October 31, 2026.

AILovable describes how its Chats feature lets a workspace-level chat agent hand work to project builder agents and receive progress back. The design records each agent's history as an append-only, forkable trajectory, and passes messages through durable inboxes that activations wake. Agents can suspend at iteration boundaries and resume on freshly deployed nodes without killing long-running runs.

Why it matters: The post details how trajectories, inboxes, and activations let agents share work and resume after deploys, useful for designing comparable agent systems.

AIJeff Dean praises Alex Ksendzovsky and team at BioComputing Co for work that lets researchers learn from computations running in real rat neurons. The quoted post from Ksendzovsky says the company is partnering with Amazon to bring the technology to customers and is taking early-access requests.

AINunchux runs MiniMax-H3 on AMD MI355X GPUs, generating 5 seconds of video in 1.3 seconds with up to 26.7x faster inference than SGLang on 8 GPUs. The stack supports streaming generation, letting users change prompts while the video plays. Free access to MiniMax-H3 through Nunchux is coming soon, with a waitlist open.

AIMike Knoop argues that deploying AI agents on top of, and within, deterministic workflows is the optimal way to deploy AI. The post is a short opinion statement and gives no specific figures, products, or benchmarks.

AILangChain released LangSmith Engine v2, an in-platform agent that scans production traces to detect agent issues and validates proposed fixes before human review. Engine v2 adds Red Teaming, currently in Private Beta for LangSmith Deployment users, which tests agents for weaknesses such as hallucinations and system-prompt violations before they reach production. Engine v2 is available in SaaS deployments for LangSmith Plus and Enterprise plans, with Self-Hosted support and BYOK for Engine coming later.

AILangChain launched LangSmith Fine-Tuning and smithtune, a CLI that turns LangSmith agent trajectories into fine-tuned models through dataset creation, training with Fireworks or Baseten, and evaluation in LangSmith. smithtune currently supports supervised fine-tuning, training models on recorded examples of good agent behavior by updating model weights. The tool lets teams train specialized models without building the data pipeline by hand.

AILangChain has launched Trajectories in LangSmith, a chronological, conversational view that aggregates human, AI, and tool messages across an agent and its subagents. Trajectories work with traces from LangChain, LangGraph, Deep Agents, OpenAI and Claude agent SDKs, and coding agents like Codex, Claude Code, and Cursor. The feature is available now on all plans in the US.