Why it matters: OpenAI's valuation reset signals how investors are re-pricing the AI boom's biggest private company.
OpenAI lowered its valuation target to $1.25 trillion after having pitched investors a higher price, according to the Financial Times.
Oracle University added two new training courses covering OpenAI Codex, the AI coding agent, aimed at developers learning to use it in their workflows.
Why it matters: Hardware enforcement of agent permissions could become a new layer for securing autonomous AI.
Nvidia introduced hardware-level controls aimed at blocking AI agents from taking unauthorized actions, according to Telecom Reseller.
Canva launched its first image generation model, available to all users. The Verge says it plans designs before generating them; Canva promises better, more customizable results at lower cost.
Why it matters: A cheap, low-latency decision endpoint broadens where OpenAI models can be embedded in agent workflows.
OpenAI's Decisions API, a low-cost endpoint for fast, focused decisions in apps, is now generally available and powered by GPT-6 Luna, with up to 10x faster decision latency, the company says.
You.com's Inference Router routes requests across open-weight AI models on demand, aiming to give enterprises reliable failover and performance for self-hosted or third-party models.
AWS released a generally available upgrade to Amazon Polly's generative text-to-speech engine, adding 14 voices across nine locales and bringing generative voices to Hindi and Finnish for the first time.
Pactum's new agent works inside Teams, Coupa and SAP Ariba to handle enterprise purchase requests. The company claims 90% cycle-time cuts without stating a baseline.
Why it matters: A major lab claims clinical superiority over physicians, raising stakes for medical AI evaluation.
Microsoft says an AI diagnostic tool it built beats generalist physicians on difficult cases, with AI CEO Mustafa Suleyman framing it as a step toward medical superintelligence. The company points to a Science study.
Dext launched an AI Connector that links its accounting and bookkeeping data to AI tools, letting firms query financial records with AI assistants.
Why it matters: Brings generative AI into Apple's mainstream productivity apps used by millions.
Apple rolls out AI updates to its iWork suite, adding a new Intelligent Assistant with content generation and editing capabilities for Pages, Keynote and Numbers, aimed at Creative Studio subscribers.
RIPL is an inference engine for Liquid AI's small foundation models that targets sub-50 millisecond latency, aimed at real-time on-device and edge agents.
Extend released a Verify API that uses AI to detect fraudulent documents, aimed at merchants and payment workflows.
Why it matters: A tiny Apache-2.0 TTS model makes on-device speech synthesis practical.
Phonon is a small text-to-speech model that converts written text into spoken audio, designed to run on-device. Gradium released its 31M-parameter weights under Apache 2.0.
Why it matters: Rerankers are core retrieval plumbing for RAG, and cheaper calibrated scores matter for agents.
Cohere Rerank 5 ships in Pro and Fast tiers with a 32K context window, Context Focus passage extraction and calibrated relevance scores; Pro scores 82.2 on BRIGHT and costs $0.05 per 1M tokens. Cohere also launched Model Vault confidential computing in public beta.
CaptivateIQ added AI agents that propose territory, quota and compensation-plan changes for administrator review. The company says beta customers updated plans 80% faster and saved 16 hours of manual work.
Traefik Labs introduces the Traefik MCP Server, which grounds AI agents in the API gateway from day 0 to day 2, letting agents interact with gateway configuration and operations.
An edit LoRA is a small add-on that teaches an image model to modify existing pictures on request. This one works with both MOSS Image Base and SFT, with weights on Hugging Face and ComfyUI custom nodes.
Why it matters: Adds a large open mixture-of-experts model tuned for medical reasoning.
Ling-3.0-flash-Sante is an open healthcare-specialized model built on Ling-3.0-flash, with 124B total and 5.1B active parameters per token, designed for medical reasoning and healthcare research.
Bojangles will roll out voice AI ordering to 100 additional drive-thru locations, expanding its use of speech automation for taking orders.
xAI opened a Grok for Startups program with an application and partner track. Bloomberg reports SpaceX is working with Apollo and Nvidia on the program; partners named include Nvidia, Databricks and Bain.
Flint-interactive is a language that lets AI agents produce interactive charts and dashboards. Readers can click a chart or filter data to explore beyond the view the author set.
Unhedged launched a service connecting AI assistants to esports, crypto and sports prediction markets, letting assistant users act on market data and events.
Nvidia's cuDSS, a library for solving sparse linear systems on GPUs, is now generally available. Partners report speedups of up to 10x, HPCwire reports.
Perfect Corp introduced a generative AI feature for its virtual eyewear try-on that removes glasses from a user's photo so shoppers can preview frames without their current pair.
Reco integrated with Cyera to prioritize AI agent risk across sensitive data and applications, giving security teams visibility into what agents can reach.
Finom, a European financial platform for SMEs, launched Finom MCP, a read-only connection letting 300,000+ business customers ask about their accounts in plain language inside their own AI assistant.
Why it matters: Gives AI agents delegated credential access while keeping secrets hidden, a template for agent identity.
1Password for Agents lets AI agents log into a user's accounts without seeing or handling passwords, with approval required for each sign-in request. 1Password's ARR passed $500 million.
Sauce Labs launched its AURA agentic release assurance offering in India, aimed at the country's fast-growing AI developer market for software testing and release quality.
Vercel Labs released fx, an open-source coding agent written in Zig designed for embedding in other applications; the code is available now.
Hystax released open-source tooling that compresses context for coding agents, making repeated build logs retrievable; in a benchmark using Codex and GPT-5.5 it cut one test bill by 22.7%.
FE fundinfo launched a Calculation Centre plus two AI capabilities to help UK investment managers meet changing production requirements under the UK CCI regime.
Why it matters: Verifying AI agent identity is becoming a prerequisite for safe agent deployments.
Identity verification firm Sumsub expanded its Know Your Agent offering with a KYA module for MCP, aiming to identify and verify AI agents interacting with systems.
HCLTech introduced an automotive software testing suite built on Nvidia technology, Just Auto reports.
Why it matters: Real-time speech translation in call centers removes language barriers for enterprise support.
ElevenLabs is targeting enterprise contact centers with live voice translation. Separately, BofA named Nvidia, Intel and Micron top AI winners as open-weight models prove expensive to run.
CrafterQ released a public API developers can use to build custom conversational AI assistants into their own products.
AgentSpot is a ThoughtSpot means for users to build and share AI agents on its analytics platform, letting business users create agents without deep engineering work.
Why it matters: A $15B raise would make DeepSeek a major China AI player and tighten Huawei's chip ecosystem.
Reports say DeepSeek has raised $15B in what could be China's largest IPO financing. SCMP reports Huawei and DeepSeek are narrowing the AI chip ecosystem gap with Nvidia, dubbed 'Jensen's nightmare'.
Consultancy Sia has launched AgentLaunch, a tool meant to let business teams build AI agents without deep technical work.
LendingTree added an AI chatbot, via a purpose-built integration, that lets borrowers explore loan eligibility, estimate borrowing costs and identify loan types before visiting its financial marketplace.
An Nvidia RTX graphics card update officially introduces Microsoft's cloud shader delivery, aimed at eliminating PC gaming stutter.
Huawei's AI glasses now let wearers pay for parking lots by scanning a QR code, adding a payment feature to the smart glasses.
Aptiv expanded its partnership with Nvidia to offer an air-cooled Thor platform for automotive AI systems.
Why it matters: Open chiplet framework could lower barriers to custom AI chip design.
Tenstorrent released the open-source beta of the Open Chiplet Atlas (OCA) Harness, a chiplet design framework that turns the OCA interoperability architecture into reusable, deployable implementations.
Why it matters: Reliability tooling for giant GPU clusters is increasingly central to AI datacenter operations.
Nvidia launches Minerva, AI operations software that enhances failure detection in giant AI data centers, aiming to catch failing components before they disrupt training or inference.
AGIBOT released a dataset aimed at teaching robots to decide what a situation needs before picking their next move, tackling tasks like restocking a minibar where grasping succeeds but ordering steps fail.
Lenovo's new Yoga laptops are aimed at AI creators, pairing Nvidia RTX Spark graphics with OLED displays, per NDTV Profit.
Why it matters: A dedicated European robot-training facility and new humanoid add capacity to the region's physical-AI push.
Humanoid robotics unicorn NEURA Robotics and TUM opened RoboGym, a new robotics training and research center in Germany, and unveiled a new robot. The facility is meant to accelerate European robot learning.
HOHOTIME is crowdfunding an all-in-one AI strength-training station with four independently adjustable resistance arms, AI coaching and a library of 860 exercises for home gyms.
OPPO launched the Find X10 series with triple 200MP cameras, agentic AI features and bigger batteries, alongside a new tablet and wireless earbuds for its connected ecosystem.
HTC launches Vivee, an egg-shaped personal AI companion device; Purdue researchers also built a wearable AI device to improve emergency allergy treatment, and the Alloop Trio offers a modular health wearable with a voice AI agent.
Why it matters: Investor nerves over AI buildout debt are starting to hit the neocloud companies renting out GPUs.
Short sellers have pocketed about $6 billion in unrealized profits as neocloud stocks fell on concerns about AI-related debt. Zadara also won Nvidia AI Cloud Ready validation, and Crusoe planned an AI cloud deployment at Paraguay's Itaipú Dam.
Gasgoo Daily reports AiMOGA unveiled its Mobo humanoid robot, alongside news that Audi named Jeff Zhang its first Chinese president and BAIC and CATL deepened a battery-swap partnership.
qBraid unveiled qBraid OS and launched hosted GH200 GPU simulators at Nvidia GTC Berlin for quantum computing development.
Why it matters: Evidence that export-banned top Nvidia chips are reaching Chinese state networks.
Israeli security firm Lava said it found 32 Nvidia Blackwell Ultra B300 chips through monitoring linked to China Telecom's network. The powerful datacenter chips are barred from sale to China.
Why it matters: A major rival buying Google TPUs weakens Nvidia's lock on AI compute and validates custom silicon.
Google's TPUs are now running AI workloads in Meta's data centers, the first time TPUs have been sold to external customers, threatening Nvidia's dominance. Reports also cover Anthropic's IPO and Meta's share issuance.
The Robotiq 2FC is a production-ready adaptive robot gripper with tactile sensing and embedded intelligence so robots can learn from physical interaction; pre-orders open now, shipping in Q1 2027.
Why it matters: Saudi Arabia's state-backed AI push adds major Nvidia compute capacity in the Gulf.
Saudi Arabia's Humain expands its strategic collaboration with Nvidia to accelerate next-generation AI infrastructure, adding an 84MW AI computing buildout in the kingdom.
Why it matters: An open 7B world-action model using only RGB could simplify robot learning stacks.
MWA-Phoenix is a 7B latent world action model for robots that uses only RGB camera input and reports leading scores on six robot benchmarks, including a real-robot test. Anyverse Dynamics released its weights openly.
Why it matters: Substrate supply is a bottleneck for AI chip packaging and datacenter buildouts.
Report says Tesla and Nvidia will invest 3.7 trillion won ($2.8B) in Samsung Electro-Mechanics for AI semiconductor substrates. Nokia and Besi also beat estimates on AI datacenter demand.
Why it matters: Rising local opposition to datacenters could slow the physical buildout of AI capacity.
Big Tech's water and power demand is sparking global resistance; investors see midterm risk to the AI trade, AI firms warn the UK has a year to head off protests, and activists interrupted Jensen Huang in Berlin.
Why it matters: Nvidia's European push and industrial agent partners shape regional AI infrastructure.
At GTC Berlin, Jensen Huang announced mass-production supercomputers and a European AI push, saying every European firm will be an AI company. A protester interrupted the keynote. Schneider Electric, Nvidia and HPE also launched industrial AI agents.
Unitree Robotics announced a World Physical AI Competition and saw its shares rise. Further details on format, prizes and eligibility were not specified in the source headlines.
Why it matters: Telecom-backed spinout signals China's push into self-evolving humanoid robot models.
Gamma Robotics, the first market-oriented company incubated by China Telecom's TeleAI, launched in Shanghai to develop self-evolving humanoid robots and whole-body models.
Robot-brain developer Qianjue launched Polibrain, a world model that predicts task-relevant state changes instead of future video frames and is adapted to quadruped, humanoid and wheeled robots.
XDA explains how changing a single llama.cpp flag let an 8GB laptop GPU run a 35B-parameter model faster than the author can read, a practical trick for running large open models on consumer hardware.
A community GGUF quantized build of DavidAU's Qwen3.8-27B Turbo-Brilliance that strips the model's refusal layer and adds 'AI construct' prompt modes. Apache-2.0 weights, ~17k downloads.
Databricks' latest release upgrades memory and infrastructure for AI agents, TechTarget reports, aiming to help agents retain context and run in production.
Oracle University added two new training courses covering OpenAI Codex, the AI coding agent, aimed at developers learning to use it in their workflows.
Canva launched its first image generation model, available to all users. The Verge says it plans designs before generating them; Canva promises better, more customizable results at lower cost.
Why it matters: A cheap, low-latency decision endpoint broadens where OpenAI models can be embedded in agent workflows.
OpenAI's Decisions API, a low-cost endpoint for fast, focused decisions in apps, is now generally available and powered by GPT-6 Luna, with up to 10x faster decision latency, the company says.
Discord's latest update renders Markdown correctly instead of showing raw symbols, a change that helps people reading agent instruction files and context shared in chat.
You.com's Inference Router routes requests across open-weight AI models on demand, aiming to give enterprises reliable failover and performance for self-hosted or third-party models.
Moonshot's Kimi AI chat assistant released version 3.2.19 for macOS, Windows and Linux.
MacWhisper, a speech recognition tool that transcribes audio locally, released version 15.8 for macOS, Windows and Linux.
The Verge assesses Meta's pitch that personal AI agents, called Muses, will shop, work and negotiate with other people's agents, questioning whether users are ready.
Why it matters: Brings generative AI into Apple's mainstream productivity apps used by millions.
Apple rolls out AI updates to its iWork suite, adding a new Intelligent Assistant with content generation and editing capabilities for Pages, Keynote and Numbers, aimed at Creative Studio subscribers.
RIPL is an inference engine for Liquid AI's small foundation models that targets sub-50 millisecond latency, aimed at real-time on-device and edge agents.
A modder reportedly jailbroke the Nintendo Switch 2, using an old game and OpenAI's Codex to build 'ns2jb', and showed Mario Kart World running with custom characters.
Why it matters: Rerankers are core retrieval plumbing for RAG, and cheaper calibrated scores matter for agents.
Cohere Rerank 5 ships in Pro and Fast tiers with a 32K context window, Context Focus passage extraction and calibrated relevance scores; Pro scores 82.2 on BRIGHT and costs $0.05 per 1M tokens. Cohere also launched Model Vault confidential computing in public beta.
AWS explains how Rescale used the AI-Driven Development Lifecycle framework with Strands Agents and Amazon Bedrock to build an intelligent test-triage agent in weeks.
An analysis of 2,050 memory entries from 80 users found ChatGPT's memory can retain personal information and psychological insights from past chats, raising questions about consent, transparency and user control.
The Linux Foundation's Agentic AI Foundation added four projects — Anthropic's Agent Skills, OpenAI's Open Responses spec, OpenEnv and OpenSandbox — bringing its first-year roster to 10.
Traefik Labs introduces the Traefik MCP Server, which grounds AI agents in the API gateway from day 0 to day 2, letting agents interact with gateway configuration and operations.
Google Cloud's API Gateway now supports SSE, WebSockets, gRPC and chunked transfer encoding in public preview, letting developers route long-lived, bi-directional and AI workloads without changing existing API specs.
A GitHub repo laying out a beginner's path through foundations, LLM engineering, agents, evaluation and production, with verified free resources and a mini-project per layer. It drew 101 stars in its first week.
Flint-interactive is a language that lets AI agents produce interactive charts and dashboards. Readers can click a chart or filter data to explore beyond the view the author set.
Amazon's Kindle Unlimited is now Kindle+, with two big changes including AI-generated audiobooks. Separately, commentary asks who will handle payments as AI agents become customers.
Manus is an AI agent app that automates workflows on your local computer. The desktop app for macOS, Windows and Linux was updated to version 2.0.10.
Google Search has dropped its Export to Docs feature for research, but users can still do the equivalent inside Gemini, according to Android Authority.
Unhedged launched a service connecting AI assistants to esports, crypto and sports prediction markets, letting assistant users act on market data and events.
TechTarget reports that CIOs' ambitious plans for AI observability are running into practical obstacles, as monitoring tools and budgets lag behind expectations.
An open-source inference engine for running large Qwen models on consumer RTX cards. v0.7.1 runs the 125B MoE Qwen3.8-Flash-Next in 64GB of RAM, and a 3-bit Qwen3.8-27B on a single 16GB card.
Nvidia's cuDSS, a library for solving sparse linear systems on GPUs, is now generally available. Partners report speedups of up to 10x, HPCwire reports.
Why it matters: A popular self-hosted inference engine adds streaming voice and routing, making local multimodal AI easier to run.
LocalAI is an open-source engine for running LLMs, vision, voice, image and video models locally, no GPU required. Version 4.12.0 redesigns the web UI and adds streaming voice responses, decision routing, image upscaling and memory controls.
A community GGUF quantized build of DavidAU's Qwen3.8-27B Turbo-Brilliance that strips the model's refusal layer and adds 'AI construct' prompt modes. Apache-2.0 weights, ~17k downloads.
A ComfyUI-ready diffusion model that turns a prompt plus ordered image, video and audio references into video with synchronized audio. Ships in BF16 and INT8 ConvRot versions based on FastH3-Omni-8-Step-V1.
Flint is an open-weight mixture-of-experts coding model with 40B total and 2.5B active parameters, built to run locally for agent coding; Recursive says much of its development was automated by AI.
Sage-A1.5 is a 35B mixture-of-experts model built for agentic tasks, released with Apache-2.0 weights on Hugging Face. It runs in Transformers, vLLM and SGLang.
Raon-1 mini is a 30B-A4B mixture-of-experts language model trained from scratch by KRAFTON AI, strong in Korean and agentic tasks, released under Apache 2.0 on Hugging Face.
A Gradio Space running an uncensored Qwen-Image-2.1 GGUF base on ZeroGPU with on-demand all-in-one LoRA adapters, letting users generate images without content filters.
Researchers introduce Future Chunk Prediction, an auxiliary training objective that pushes autoregressive video diffusion models to model how scenes evolve rather than rely on temporal redundancy for local continuity.
DONIC is a research model that learns generalist two-handed dexterous manipulation from human demonstrations, aiming to cover the contact-rich bimanual behaviors common in real life that single-hand trackers miss.
A paper examines whether gains from RAG components like chunking, reranking, query augmentation and agentic verification hold up when combined, rather than evaluated individually.
Quant-Arena is a new benchmark testing whether LLM agents can run the full quant research loop: propose a market hypothesis, code it as a signal, backtest it and decide what to keep, with results judged beyond the research period.
A GitHub repo laying out a beginner's path through foundations, LLM engineering, agents, evaluation and production, with verified free resources and a mini-project per layer. It drew 101 stars in its first week.
LocalAI is an open-source engine for running LLMs, vision, voice, image and video models locally, no GPU required. Version 4.12.0 redesigns the web UI and adds streaming voice responses, decision routing, image upscaling and memory controls.
A Python project that implements a general agent runtime from scratch, with an agent loop, permissions, MCP, sub-agents, context folding and self-evolving skills. It drew 103 stars in its first week.
A self-hosted TypeScript tool that converts written requirements into UML diagrams, test cases and editable Word specs with traceability, using your own AI model. Aimed at course and capstone projects.
Pydantic AI is a Python framework for building agents, realtime voice apps and image generation across models. Version 2.59.0 fixes a high-severity HTML conversion issue in the web_fetch tool.
Canva launched its first image generation model, available to all users. The Verge says it plans designs before generating them; Canva promises better, more customizable results at lower cost.
Discord's latest update renders Markdown correctly instead of showing raw symbols, a change that helps people reading agent instruction files and context shared in chat.
Moonshot's Kimi AI chat assistant released version 3.2.19 for macOS, Windows and Linux.
MacWhisper, a speech recognition tool that transcribes audio locally, released version 15.8 for macOS, Windows and Linux.
The Verge assesses Meta's pitch that personal AI agents, called Muses, will shop, work and negotiate with other people's agents, questioning whether users are ready.
Apple rolls out AI updates to its iWork suite, adding a new Intelligent Assistant with content generation and editing capabilities for Pages, Keynote and Numbers, aimed at Creative Studio subscribers.
A modder reportedly jailbroke the Nintendo Switch 2, using an old game and OpenAI's Codex to build 'ns2jb', and showed Mario Kart World running with custom characters.
An analysis of 2,050 memory entries from 80 users found ChatGPT's memory can retain personal information and psychological insights from past chats, raising questions about consent, transparency and user control.
Amazon's Kindle Unlimited is now Kindle+, with two big changes including AI-generated audiobooks. Separately, commentary asks who will handle payments as AI agents become customers.
Manus is an AI agent app that automates workflows on your local computer. The desktop app for macOS, Windows and Linux was updated to version 2.0.10.
Google Search has dropped its Export to Docs feature for research, but users can still do the equivalent inside Gemini, according to Android Authority.
Unhedged launched a service connecting AI assistants to esports, crypto and sports prediction markets, letting assistant users act on market data and events.
PCMag compares free AI image editors LightCraft and PhotoCraft against Photoshop to see whether they can handle everyday photo editing tasks.
Perfect Corp introduced a generative AI feature for its virtual eyewear try-on that removes glasses from a user's photo so shoppers can preview frames without their current pair.
Pickle, an AI agent app from a Silicon Valley startup, is shaking up Korea's tech scene, according to Herald Business.
Kokon is a no-settings desktop photo editor for the open Qwen-Image 2.1 model, letting users generate images without ComfyUI nodes. Version 0.10 adds Prompt+, which writes detailed multilanguage prompts, plus beta custom LoRAs.
Time reports older Americans use Meta's Muse AI agent for tasks and companionship, while privacy experts flag risks from the agent's access to personal data.
Impact Factory released a marketing plug-in that lets brands run campaigns and connect with customers inside ChatGPT and other leading AI assistants.
Google released Gemini Spark in India, pitched as a 24/7 AI agent that runs tasks in the background, including while the user sleeps.
Granola is an AI notepad that transcribes back-to-back meetings so users stay present instead of taking notes. Version 1.15.11 brings a new Duo feature to the iPhone and iPad app.
Code in Google One suggests Google is developing a 'Connected accounts' feature to share storage and AI perks across accounts without a family group.
Zoom announced a unified context layer and an AI agent suite, BigGo Finance reports. Oppenheimer upgraded the stock to Outperform with a $130 target.
DeepCell is a web app and agent plug-in that produces documents, spreadsheets and slides with sources for deadline tasks like modeling, budget forecasts and monthly reports, with linked edits.
Radaric is an AI feed reader that pulls RSS, Telegram, X and YouTube into one feed and learns from what you approve or reject, merging repeat coverage; from $9/mo.
Bits is an open-source personal agent running on your own machine that never sleeps: a background daemon runs scheduled routines and resumes tasks after crashes, with memory in editable Markdown and risky actions gated by a reviewer.
XDA explains how changing a single llama.cpp flag let an 8GB laptop GPU run a 35B-parameter model faster than the author can read, a practical trick for running large open models on consumer hardware.
Databricks' latest release upgrades memory and infrastructure for AI agents, TechTarget reports, aiming to help agents retain context and run in production.
Oracle University added two new training courses covering OpenAI Codex, the AI coding agent, aimed at developers learning to use it in their workflows.
OpenAI's Decisions API, a low-cost endpoint for fast, focused decisions in apps, is now generally available and powered by GPT-6 Luna, with up to 10x faster decision latency, the company says.
You.com's Inference Router routes requests across open-weight AI models on demand, aiming to give enterprises reliable failover and performance for self-hosted or third-party models.
RIPL is an inference engine for Liquid AI's small foundation models that targets sub-50 millisecond latency, aimed at real-time on-device and edge agents.
Cohere Rerank 5 ships in Pro and Fast tiers with a 32K context window, Context Focus passage extraction and calibrated relevance scores; Pro scores 82.2 on BRIGHT and costs $0.05 per 1M tokens. Cohere also launched Model Vault confidential computing in public beta.
AWS Machine Learning · 5h
#
AWS explains how Rescale used the AI-Driven Development Lifecycle framework with Strands Agents and Amazon Bedrock to build an intelligent test-triage agent in weeks.
The Linux Foundation's Agentic AI Foundation added four projects — Anthropic's Agent Skills, OpenAI's Open Responses spec, OpenEnv and OpenSandbox — bringing its first-year roster to 10.
Traefik Labs introduces the Traefik MCP Server, which grounds AI agents in the API gateway from day 0 to day 2, letting agents interact with gateway configuration and operations.
Google Cloud's API Gateway now supports SSE, WebSockets, gRPC and chunked transfer encoding in public preview, letting developers route long-lived, bi-directional and AI workloads without changing existing API specs.
Flint-interactive is a language that lets AI agents produce interactive charts and dashboards. Readers can click a chart or filter data to explore beyond the view the author set.
TechTarget reports that CIOs' ambitious plans for AI observability are running into practical obstacles, as monitoring tools and budgets lag behind expectations.
An open-source inference engine for running large Qwen models on consumer RTX cards. v0.7.1 runs the 125B MoE Qwen3.8-Flash-Next in 64GB of RAM, and a 3-bit Qwen3.8-27B on a single 16GB card.
Nvidia's cuDSS, a library for solving sparse linear systems on GPUs, is now generally available. Partners report speedups of up to 10x, HPCwire reports.
ModelOp expanded Microsoft and Amazon integrations to help enterprises control AI agent sprawl across their deployments.
Shopify Engineering · 9h
#
Shopify Engineering describes Remote Dev, an internal environment that lets AI coding agents build and test apps in parallel without overloading developers' laptops.
Vercel Labs released fx, an open-source coding agent written in Zig designed for embedding in other applications; the code is available now.
Hystax released open-source tooling that compresses context for coding agents, making repeated build logs retrievable; in a benchmark using Codex and GPT-5.5 it cut one test bill by 22.7%.
Automation platform Workato launches its AI Agent Builder in Singapore, giving Asian developers and businesses access with 50,000 free starting credits.
HCLTech introduced an automotive software testing suite built on Nvidia technology, Just Auto reports.
CrafterQ released a public API developers can use to build custom conversational AI assistants into their own products.
AgentSpot is a ThoughtSpot means for users to build and share AI agents on its analytics platform, letting business users create agents without deep engineering work.
Consultancy Sia has launched AgentLaunch, a tool meant to let business teams build AI agents without deep technical work.
Google's Gemini 3.6 Flash model has appeared in Antigravity, the company's agent-first coding IDE, reportedly after several days of testing.